ML/1950 6 fundamenta informatica io press optim rough set model institut comput scienc technolog poland abstract rough set methodolog base concept approxim construct background knowledg repres inform system applic partial knowledg approxim concept parametr famili concept approxim built tune paramet approxim chosen variabl rough set model approxim space paper follow approach gener approxim space discuss rough set model base approxim space uncertainti function rough element approxim space parametr proper applic model data set make optim paramet discuss basic properti mention model strategi paramet optim present notion rough relat optim paramet base relat defin condit attribut illustr present method real medic data set includ introduct rough set theori approach knowledg discoveri incomplet data rough set approach process incomplet data base lower upper approxim rough set defin pair crisp set approxim approxim subset mention subset defin respect inform roughli defin finit set call object set attribut valu standard rough set approach introduc optim rough set model pawlak relat symmetr transit object defin relat attribut valu relat construct assum exist equal relat attribut valu object indiscern view object inform inform object gener indiscern relat standard rough set model lower approxim subset defin union class fulli includ hand upper approxim defin union class intersect real data set obtain measur mistak made collect data situat notion approxim definit restrict extens direct variabl rough set model indiscern relat emploi order defin approxim set approxim relat investig relat approxim motiv theoret practic point view mention exampl equal approxim fundament gener rough set approach base similar relat approxim equal relat set attribut rough set method control process requir function approxim problem interest subset relat defin term inform paper discuss approach base gener approxim space introduc investig combin model extens indiscern relat extens standard definit approxim standard rough set model approach unifi case consid modif origin approxim space definit concern call uncertainti function inform repres attribut vector set object similar attribut vector vector creat set object vector creat indiscern class relat case relat modif approxim space definit introduc gener rough membership function assum answer question object belong object set answer question includ primit notion rough function rough membership function paper organ section properti approxim gener approxim space discuss present result approxim relat section construct approxim space discuss search optim approxim space tune paramet inform granul section present result experi real data set diabet optim rough set model approxim approxim space section present gener definit approxim space introduc toler base rough set model variabl rough set model approxim space subsect defin basic notion section paper set denot set subset definit approxim space system set uncertainti function rough function condit impli tripl impli tripl empti set collect subset subset ffl ffl ffl satisfi condit versu rough condit formul rough object necessarili set observ assum satisfi sixth condit condit bridg standard set theoret rough call function sri standard rough function sri function rough variabl rough set model mention condit rough optim rough set model definit satisfi function defin approxim space notion defin lower upper approxim definit parametr approxim space subset lower upper approxim defin low fx fx recal notion posit region qualiti approxim classif case gener approxim space definit approxim space fx classif object posit region classif fx respect approxim space defin po fx low qualiti approxim classif fx approxim space defin fl fx qualiti approxim classif coeffici express ratio number classifi object number object data tabl coeffici classif approxim space base entropi properti approxim rough defin subsect list properti approxim gener approxim space present definit algorithm check rough intern undefin approxim space defin set equal respect rough proposit assum sri standard rough show properti sri sri low sri sri low sri sri optim rough set model sri sri low sri low low sri low low sri impli sri low sri impli sri low sri sri low sri low low sri low sri sri analog standard rough set theori defin type roughli low sri sri intern low sri sri extern low sri sri total low sri sri intuit mean classif roughli decid element belong intern decid element belong unabl decid element belong extern decid element belong unabl element belong total unabl decid element belong algorithm check properti set time card observ properti low sri pp sri obtain intern extern algorithm case optim rough set model approxim space rough relat distinguish direct research relat approxim list exampl properti rough relat present relationship rough relat modal logic investig author refer upper approxim relat program respect indiscern relat ind treat composit ind ffi ffi ind special symbol lower approxim introduc properti relat approxim gener approxim space present relationship rough set algebra relat investig relationship rough relat problem object rank present shown classic rough set approxim base indiscern relat account ordin properti consid criteria drawback remov rough approxim prefer relat grade domin relat subsect discuss approxim relat respect rough simplic present binari relat approxim nonempti set denot project relat axi fx rg definit rough function base cardin project defin relat describ intuit mean approxim approxim space standard lower approxim low sri relat object connect lower approxim object relat obtain restrict definit lower approxim rough pair lower approxim low pair obtain similar interpret upper approxim respect introduc rough pair upper approxim pair proposit lower upper approxim condit optim rough set model low sri low sri low low sri low sri give illustr proposit replac equal observ equal cartesian product set binari relat definit uncertainti function rough tabl sri lower upper approxim approxim space tabl low sri low low proposit time complex algorithm comput approxim relat equal search approxim space section problem search adequ approxim space search proper approxim space crucial difficult task relat algorithm synthesi base approxim space optim uncertainti function approach search uncertainti function base assumpt metric attribut valu relat obtain attribut valu metric reflex review metric defin attribut valu present exampl metric tabl differ metric distanc function nomin attribut version vdm weight optim rough set model scheme normal defin distanc valu attribut vdm sri sri number sri standard rough fx ig fx vg distanc measur valu consid closer similar classif attribut color valu green applic identifi object red green consid closer red blue correl appl distanc function directli continu valu potenti uniqu approach problem vdm continu attribut present distanc function real integ valu diff jv max min max min maximum minimum attribut search optim uncertainti function ffl class uncertainti ffl optim criterion ffi distanc function attribut set valu attribut defin uncertainti function fa ffi function defin real number special exampl uncertainti function deriv literatur strict weak indiscern relat consid defin kind uncertainti function case natur relat defin defin real valu attribut ffi ja ff fi ff fi real number global uncertainti function set attribut defin intersect exampl local global uncertainti function optim rough set model method search paramet proper uncertainti function discuss genet algorithm appli search adequ uncertainti function type optim criterion base maxim function combin quantiti quantiti base qualiti approxim classif fl fx specif quantiti defin relat introduc rough relat measur ffl sri ffl ffl quantiti base result test optim simplest case combin quantiti weight fl fl fx weight sri weight test test weight fl weight weight test weight fl weight weight test case weight test part object function introduc prevent shrink posit region partit part object function respond number connect interest connect object function find kind balanc enlarg prevent shrink posit region po fx object chang larger start threshold properti find valu creat list threshold check combin threshold find long worst number combin equal card card show heurist close optim solut reason time genet algorithm applic method handwritten numer recognit report optim approxim space tune paramet inform granul optim approxim space paramet strongli relat inform granul construct subsect illustr gener idea inform granul construct context approxim space paramet descript granul type relat symbol paramet logic interpret paramet tune optim set paramet formula ff defin optim rough set model relat structur set close satisfactori concept approxim process dt tabl class set condit attribut fa dt defin finit model formula ff ff approxim space paramet vector parametr formula defin famili subset object set tune process lead paramet vector fx ff satisfactori approxim concept chosen equal set object extens obtain ad object call test object inform vector element cartesian product set requir pose inform granul import applic mention ffl minimum descript length principl obtain short sum length descript model formula ff ff constraint ff describ suffici close class ffl adapt flexibl dt extend dt ad tune paramet chang structur formula ff ff model ffl finit model formula ff ff class problem case neural network gener approach descript inform granul tabl dt ffl finit model ffl formula ff ff ffl close measur object ffl optim criteria tune paramet vector semant descript inform granul fix paramet vector vector set ff realiz ff close object set express close sri discuss exampl search optim approxim space tune paramet inform granul optim rough set model dt hypothet medic data tabl fx set condit attribut fx fx fx nog fx sex infect defin approxim space paramet ffl fy ffl sri sri lower upper approxim low fx fx low fx fx paramet obtain low fx fx low fx fx assign tabl dt approxim space finit relat structur sex infect set condit attribut dt constant lower approxim find suitabl optim rough set model ff low sex infect ff low infect ff low infect ff low infect set ff low equal low set ff low equal low tune process extract approxim space lead optim paramet elementari granul defin tune process simpl formula combin boolean connect match lower upper approxim optim criterion respect approxim space paramet base good predict accuraci test set object optim approxim space search adequ formula adequ languag process handwritten digit recognit problem formula proposit logic similar problem discuss formula order logic search optim uncertainti function diabet data tabl section real data set children diabet studi data depart children medic school poland discuss search optim uncertainti function data set aspect evalu patient suffer depend diabet ag famili criteria metabol type appli presenc absenc rule classif subsect sketch basic method experi problem relat practic applic rough set method set attribut determin simplifi suffici subset attribut origin problem find reduct gener approxim space transform problem find prime implic monoton boolean function tabl defin subset approxim space sri definit recal notion object relat reduct optim rough set model definit subset call rel reduct object fy fy proper subset ae condit satisfi detail definit properti type reduct gener approxim space call classif gener reduct order comput reduct comput experi reduct comput exhaust calcul method find reduct comput prime implic boolean function comput effort li reduct comput rule comput time conceptu overlai reduct tabl comput read attribut valu attribut reduct relat object construct selector form paramet definit uncertainti function classif perform algorithm present object algorithm scan rule set determin rule fire rule direct conflict content inform vector fire rule major voter appli mean domin data set suggest rule class elect process fire rule perform order resolv conflict rank accumul vote fire rule accumul number vote defin certainti coeffici class class largest certainti coeffici select ti resolv major voter algorithm result experi tabl repres clinic experi treatment diabet construct condit attribut express patient data attribut data preliminari discret data preliminari discret suggest medic expert case obtain rule inform medic expert selector obtain rule necessarili exact correspond norm medic doctor decid make preliminari discret suggest medic expert attribut valu present tabl optim rough set model symbol attribut attribut sex ag diagnosi durat diabet famili therapi type kit iit system infect code tabl construct valu attribut code natur number discret code obtain set parametr approxim space construct code tabl construct approxim space asa sri card rough sri defin standard condit attribut ffi ffi diff fa vdm fa mean appli diff metric attribut code condit attribut differ metric appli mean sri sri ja set attribut intersect check threshold call essenti set threshold essenti minim respect set satisfi condit attribut obtain set essenti optim rough set model number essenti threshold combin greater present select result sort descend order result leav test includ qualiti approxim classif fl standard rough sri fl sri est observ result result obtain threshold equal correspond standard rough set approach result experi present graphic figur gener filter rule present paramet accept medic expert test figur select result standard rough set relat defin approxim space paper discuss gener approxim space notion cover extend rough set toler base rough set model variabl rough set model approach base simultan extens optim rough set model indiscern class notion extens relat indiscern class approxim set point role search optim paramet approxim space connect search tune paramet inform granul space essenti coeffici describ approxim space huge heurist valuabl genet algorithm optim approxim space good solut found reason time adequ approxim space import extract law tabl diabet data set drawn real life medic problem result analysi extract law accept medic expert acknowledg author professor skowron critic remark work support grant state scientif research technolog rector grant research program european union refer nguyen nguyen skowron synthesi rule object incomplet rough set pp fuzzi extens rough set intern confer rough set current trend june lectur note artifici intellig pp system softwar implement rough seventh intern workshop intellig inform june pp rough set algebra incomplet rough set pp rough approxim prefer relat comparison rough set knowledg discoveri case studi softwar pp fuzzi similar relat basi rough intern confer rough set current trend june lectur note artifici intellig pp optim rough set model kim kim handwritten numer charact classif toler rough manuscript learn rule similar base rough rough set knowledg discoveri case studi softwar pp select object toler rough set poster session ninth intern symposium methodolog intellig june pp nguyen nguyen pattern extract fundamenta informatica pp nguyen skowron discoveri data pattern applic decomposit classif rough set knowledg discoveri case studi softwar pp orlowska inform lectur note comput scienc pp pawlak rough bulletin polish academi technic scienc vol pp pawlak rough set theoret aspect reason kluwer academ pawlak skowron rough membership advanc theori john wilei pp skowron rough lectur note artifici intellig berlin pp skowron adapt calculu intern pp skowron synthesi system data rough set data mine analysi kluwer academ pp skowron learn toler relat boolean automat featur extract data fourth intern workshop rough fuzzi machin novemb pp skowron approxim theori discret fundamenta informatica pp skowron approxim rough fuzzi set knowledg springer london berlin pp optim rough set model skowron gener approxim intern workshop rough set soft san novemb pp skowron toler approxim fundamenta pp skowron construct inform world congress scientif model appli august vol artifici intellig comput pp skowron inform granul approxim seventh intern confer inform process manag uncertainti juli pp strict weak indiscern object quantit attribut overlap foundat comput vol pp properti applic rough ic research report intern workshop intellig inform june pp system base toler rough fourth intern workshop intellig inform june pp attribut discoveri rough principl data mine knowledg european june lectur note artifici intellig springer pp approxim reduct rough set knowledg discoveri case studi softwar heidelberg pp rough relat rough set knowledg discoveri methodolog heidelberg pp urban applic rough set base data mine techniqu diabet seventh intern workshop intellig inform june pp approxim space extens rough set intern confer rough set current trend june lectur note artifici intellig pp approxim reduct intern confer inform process manag uncertainti juli pp optim rough set model yao wong lin review rough set rough set data mine analysi kluwer academ pp yao gener pawlak approxim intern confer rough set current trend june lectur note artifici intellig pp rough modal journal appli vol pp urban applic rough set theori diabet preval preliminari commun wilson martinez improv heterogen distanc journal artifici intellig vol pp variabl rough set journal comput system vol pp 
DB/1920 3 progress report disjunct deduct system dlv thoma eiter nicola leon matei gerald pfeifer francesco scarcello institut fur tu wien austria institut fur universitat dei universita della itali abstract dlv deduct base disjunct logic offer advanc kr formal system develop end technisch universitat wien ongo project fund austrian scienc fund recent comparison shown dlv nowadai implement disjunct logic program major strength dlv advanc knowledg model featur kernel languag extend disjunct logic program strong negat la gelfond integr languag diagnost reason suitabl interfac dlv user util base relat store extern commerci system paper overview dlv system describ recent advanc implement recent implement increment techniqu program relat join optim strategi relev deduct applic techniqu suitabl extend dlv effici instanti nonmonoton program benchmark problem domain report give feel current system perform introduct deduct extend relat system inferenti power logic program studi major activ research partial support scienc project queri system disjunct deduct italian murst project cooper advanc formal system nonmonoton reason address correspond author half eighti import theoret studi led implement number deduct system support logic program extens queri languag idea incorpor disjunct deduct languag stimul renew interest deduct call disjunct deduct suit perform nonmonoton reason task strongli field artifici intellig research semant complex disjunct deduct interest studi express power shown concret real world problem repres deduct repres problem mention strateg compani busi domain describ section famou travel salesman problem field problem find minimum size kei relat kei minimum number domain gener consid power tool reason knowledg shortag actual implement paper system dlv develop tu wien project queri system disjunct deduct major strength dlv advanc knowledg languag support kernel languag extend disjunct logic program strong negat la gelfond integr languag diagnost reason suitabl interfac dlv user access relat store extern commerci dlv advanc interfac relat section describ languag support system address knowledg show sophist queri easili elegantli dlv section address implement issu present architectur dlv describ optim techniqu recent incorpor final report benchmark give feel current system perform previou version includ optim techniqu present compar system inform download visit project homepag languag dlv extend disjunct logic program nativ interfac system offer full power intern languag disjunct logic program extend integr constraint strong negat presenc construct make languag easili repres wide rang problem natur highli declar dlv rule lm ln atom negat predic constant variabl convent prolog constraint rule dlv program pair ic set dlv rule ic set constraint stabl model consist answer set satisfi constraint ic present sampl drawn encod nativ languag dlv give idea languag knowledg reason problem benchmark evalu perform system find citi citi reachabl classic deduct provid singl version repres fact form assign node color adjac node color classic problem encod easi natur mean disjunct disjunct rule nondeterminist color node constraint enforc choic legal find prime implic proposit cnf formula ic liter variabl xm recal prime implic consist conjunct liter subformula easili prime implic exist satisfi find prime implic subsum satisfi problem dlv natur simpl true negat languag plai kei role liter translat dlv program nc easi stabl model correspond prime implic strateg compani problem kr problem busi domain result kr problem hold produc good compani jointli control compani compani constraint good compani sold control hold transact compani belong strateg minim set compani satisfi constraint set express natur mean produc product produc compani control jointli control adopt product produc compani compani control compani problem find set strateg compani brave fact note problem express fix normal logic program uniformli collect produc control np brave cautiou reason frontend brave cautiou reason extens nativ interfac addit user specifi essenti bodi rule question ground liter brave reason frontend invok command line option queri evalu true stabl model system brave evidenc return stabl model wit answer brave cautiou frontend invok system answer true stabl model case neg evidenc stabl model wit falsiti frontend possibl subset queri intern translat nativ system languag standard complet resort current work draft system automat invok sql frontend input file carri extens sql column includ construct creat connect paramet datalog predic column sql datalog schema grammar describ queri express handl sql select union item repres list separ express mean express word capit letter refer construct trail observ queri express gener definit part queri part canon list materi datalog schema union select select minor queri translat element print result rule repres definit part rule correspond queri part queri express rule intern predic queri part gener gener queri consist connect set oper implement featur languag defin set oper explicit addit comparison predic support add arithmet predic diagnost reason dlv kind diagnosi abduct diagnosi frontend reiter diagnosi frontend studi peirc import kind wide applic area comput recogn import principl reason abduct diagnosi deal problem find explan observ base framework repres logic program roughli abduct modu observ abduct conclud explan diagnost problem formal tupl hyp set ground atom ob set ground liter manifest logic program explan subset hyp satisfi impli ob brave exist stabl model liter ob true diagnost problem explan accord occam principl parsimoni state explan simpler minim criterion wide criterion subset state solut prefer solut ae accept solut restrict minim explan consid criterion work assumpt singl error consid elementari consist singl relat scheme consist attribut student function depend teacher class class room room teacher room student find kei relat model problem abduct diagnost problem kei kei theori kei teacher class class room room teacher room student cover encod function depend state cover relat attribut impli attribut relat set observ diagnosi cover relat minim diagnosi kei vice versa minim diagnosi kei relat singl error correspond singl attribut kei frontend abduct diagnosi support gener subset minim diagnosi singl failur diagnosi mode invok command line option posit theori diagnost theori obei syntax section hypothesi list atom separ dot store file carri extens hyp diagnost frontend work read disjunct rule guess diagnosi candid gener observ constraint forbid gener inconsist diagnosi rule ad prune solut ground subsequ mg report stabl diagnosi return frontend diagnosi reiter work similarli command line option implement issu handler model checker rule graph ground program diagnost reason depend graph edb oracl file system ground intellig edb extend dlp model gener queri processor translat translat fig architectur system system architectur outlin gener architectur system depict figur heart system li queri processor control execut entir system collabor integr frontend perform input gener queri processor read possibl input program hand rule graph split subprogram relat provid oracl ascii text subprogram submit intellig ground effici gener subset ground input program stabl model full smaller gener queri processor invok rule graph gener partit program model gener model enabl modular evalu program yield tremend speedup model gener start gener candid stabl model invok model checker verifi stabl model control return queri perform possibl invok mg model detail intellig ground model checker found effici instanti present intellig ground modul dlv system aim ig evalu stratifi program gener instanti disjunct compon input program disjunct order effici evalu stratifi program princip techniqu classic deduct relat optim theori improv version gener evalu techniqu linear rule obtain effici evalu rule bodi case larg relat join optim techniqu end input program normal ig complet evalu program modul emploi program singl stabl set atom return instanti procedur input program disjunct instanti procedur complet evalu program techniqu mention effici gener instanti nonmonoton part program aspect crucial number gener ground time gener ground size ground gener import strongli influenc comput time modul system slower instanti procedur gener smaller ground prefer faster gener larg ground time gain comput time main reason larg input atom rule instanti atom lead combinatori reason program atom deriv instanti render applic rule present algorithm gener ground instanc rule atom possibl deriv denot rule part associ depend graph gp node gp predic head rule gp arc node node iff exist rule simpl optim emploi depend graph gp result gp maxim strongli connect compon singl node note node gp set node gp optim realiz techniqu relat deduct optim mention program node graph gp denot rule set rule st predic occur exit rule remain set rule predic rule total normal stratifi bodi liter defin total rule atom total rule defin total recal assum variabl rule order instanti rule instanti uniqu determin instanc defin ground wrt set ground atom nf denot set ground instanc st nf set comput function evalu end section figur outlin algorithm comput ground program set ground rule set ground atom deriv ground rule empti stabl model denot instanti set nf remov node gp incom arc gener instanc rule defin predic call procedur procedur gener ground gener rule neg total liter true wrt add nf atom possibl deriv remov posit liter neg total head bodi empti insert simplifi ad order comput function evalu proce match atom atom nf bind free variabl case relat join oper exit rule set irrelev linear optim techniqu bodi atom match atom atom match atom nf improv gener techniqu effici heurist start posit liter predic occur infrequ nf variabl find bodi liter start match atom order posit liter bodi cardin ground occurr nf number share variabl posit liter uniqu variabl end rule cardin ground occurr nf small join oper rest bodi liter cartesian product describ function evalu proce exit rule linear rule liter match match bodi liter note variabl due previou step match atom nf bodi comput match match liter bodi add ground instanc match atom match atom nf bodi liter ground instanc deriv function evalu exit return set backtrack previou liter match atom nf function evalu proce similar mark bodi liter time time match bodi liter fail case exit rule linear function exit return set unmark current mark bodi mark bodi liter step case exit rule linear rule mark bodi liter match atom bodi liter left mark bodi liter match atom nf bodi liter mark bodi liter match atom nf classic gener techniqu make differ bodi liter left mark bodi gener duplic ground instanc improv avoid gener ground rule function evalu exit match bodi liter fail bodi liter mark effici rule instanti rule repeatedli instanti nf remain unchang sourc process gp empti time pick sourc gp deriv ground instanc gener ground instanc sourc process optim gener ground rule head atom possibl deriv note normal stratifi ground empti bodi ground rule empti head atom ad procedur safe var var var begin nf gp depend graph gp remov node gp incom exit nf repeat nf end end procedur safe var var var begin instanc nf nf remov posit liter remov neg liter st total fh end end end fig algorithm comput instanti program program edb predic instanti comput set ground atom ground program evalu node fpg yield upper left rule nf evalu node fqg upper rule nf node ftg rule yield ad remain rule yield note instanc instanti gener rule theorem safe output paramet return stabl stratifi stabl model benchmark section report result dlv system run time benchmark problem domain give feel current perform system problem consid reachabl strateg compani fig graph simplex graph result experi displai figur run sun ultra solari version compil map citi provid stanford instanc gener vari paramet instanc pair citi distanc directli connect result displai tabl figur left diagram show number direct connect citi size number graph size direct total execut time graph size edg total execut time result reachabl number edg total execut time number edg total execut time result graph result simplex graph graph total execut time result strateg compani fig result tupl input show total time system comput output relat refer size number output number direct connect graph high number transit connect result instanc consist direct yield transit process minut graph perform direct connect tupl dlv approxim minut encourag ongo work advanc data structur effici support applic consid reachabl ground forward chain benchmark primari applic domain dlv graph consid class graph simplex graph figur class evalu perform deduct smodel result show dlv effici scale advanc nonmonoton infer mechan exploit strateg gener instanc compani product compani control compani member actual number compani uniform randomli chosen averag control relat compani result dlv interest problem report number control relationship coupl compani roughli evalu fast hard instanc high comput complex problem similar test prove understand complex natur problem refer niemela research project disjunct logic program ai zicari integr data model program paradigm acmsigmod intern page atlant gottlob default logic queri languag ieee transact knowledg data zaniolo ldl system prototyp ieee transact knowledg data truszczynski default reason system intern confer principl knowledg reason scarcello dlv model gener applic frontend workshop logic program research report page septemb munchen leon abduct logic semant complex theoret comput decemb mannila disjunct datalog acm transact septemb scarcello deduct system nonmonoton reason intern confer logic program nonmonoton reason number lectur note ai springer scarcello kr system progress comparison benchmark technic institut fur tu gelfond lifschitz classic negat logic program disjunct gener toni abduct logic program journal logic kaka mancarella updat abduct page knuth stanford platform combinatori comput acm leon rullo system integr reason invit paper ieee intern confer tool artifici novemb scarcello disjunct stabl unfound fixpoint semant comput inform june niemela simon effici implement stabl model semant joint intern confer symposium logic page sept peirc abduct induct philosoph write chapter peng abduct infer model diagnost problem solv symbol comput artifici intellig ross deduct system confer manag page pool explan architectur default abduct reason comput sudarshan coral relat logic vldb british reiter theori diagnosi principl artifici system reason disjunct deduct intern workshop deduct approach inform system dbl work foundat ansi apr temporarili 
HCI/1748 4 design histori mechan collabor educ simul catherin ann gari richard ben interact institut advanc comput center engin learn system comput scienc colleg md abstract review past event domain videotap flight data record provid invalu technolog sport coach aviat engin provid learner readabl record action monitor reflect experi experi facilit activ collabor learn commun learn histori student make effect digit librari word process design electron perform support web navig paper describ design space discuss challeng implement learn histori present guidelin creat effect design tradeoff histori record paper present implement learn histori engin learn environ call simpl process learn case semiconductor fabric report earli user evalu learn histori implement simpl simul introduct footbal player review videotap recent game search exampl plai work bridg coach analyz strategi offer guidanc tactic trainer extract exemplari plai futur instruct session retir player add comment dramat review activ place air safeti inspector examin flight data record plane crash wrong design pilot learn prevent disast care multipl commentari expert appli videotap flight data record provid valuabl technolog assist review past capac easili record activ benefit learner comput entir session peer mentor analyz student work educ design tool conjectur learn histori encourag metacognit encourag student monitor behavior reflect progress user comput system experienc current state graphic user interfac offer undo capabl user walk backward retrac modest function greatli expand benefit captur manipul histori user action aid learn environ give student access record action review previou step understand correct earlier step replai entir histori save histori review consult peer mentor search histori individu step sequenc send annot histori email peer mentor combin histori collabor project demonstr tutori guid student complex procedur creat macro repeat sequenc exploratori research histori benefit learner directli wide rang domain student improv skill digit librari word process design electron perform support web navig hill studi web user show url visit web navig benefit greatli power histori tool version web navig histori includ gener limit improv version commerci simpl design controversi abound strategi produc list network appar approach web visit histori demonstr benefici greenberg histori facilit collabor typic involv student review build contribut student product consist text draw easi tradit technolog email bulletin board dissemin product review project review involv complex sequenc action learn conduct physic experi emerg difficult review work sequenc action context rich histori record invalu learn histori engin simul sequenc time action dramat effect simul fabric chemic process learn histori engin simul main focu rest paper approach consist make record action learner differ approach assist learn monitor analyz user action goal categor user behavior prescrib lesson instructor student emploi tradit approach make histori data complementari paper describ design space challeng implement learn histori describ implement learn histori semiconductor fabric implement engin learn environ call simpl process learn develop maryland final report earli user evalu prototyp goal research learn histori mechan record suffici inform user action system state enabl review previou action impact accomplish requir effici intern user action accompani user interfac present inform level function requir care proper softwar order user relationship action simul outcom desir featur robust learn histori mechan adjust replai search macro figur histori record user manipul control simul left action record comment ad record object proper softwar engin approach promot learn histori persist data structur exist independ process environ creat histori email peer mentor post search challeng class submit action sequenc accomplish goal fewer fewer resourc log tradit system provid suffici inform implement gener interact gui environ difficult adjust replai speed videotap view learn histori replai normal faster slower step critic sequenc backward step time benefit capac summari long sequenc action zoom detail figur mail histori question peer mentor annot annot critic compon commun histori learn environ gener automat annot kei event accomplish user highlight comment action pose answer relat element learn environ refer valv valv highlight graph show effect valv close annot histori made histori complement audio record record user comment describ problem step led enrich provid access relat tool guidanc annot mechan instructor creat demonstr tutori conveni carri set instructor demonstr common emerg common mistak instructor annot augment histori kei point time student replai session annot make time provid guidanc student direct student ey event depend group interact annot privileg vari identifi student permit add comment minimum requir instructor edit histori hide unwant student prohibit replai histori enabl repair problem capac altern conveni singl action yield complex anticip modifi part histori invalid action produc result permit depend applic analysi histori search search facil enabl student find pattern lengthi simul sequenc search facil enabl instructor find student achiev fail achiev pattern work common search problem complex histori pattern captur state relationship larg number heterogen control macro advanc learn histori system includ capac creat execut macro user carri action sequenc add iter condit creat program numer simul carri rapidli semiconductor manufactur student vari temperatur pressur larg rang automat run hundr simul impact time complet macro facil rich possibl modular strategi addit featur learn simplifi strategi wise extens experienc learner featur learn benefici expert facilit collabor analysi optim model simul li heart system engin featur macro capabl autom complex sequenc design iter optim learn histori design issu design tradeoff abound develop tool learn histori histori easier review inform complet record event occur session provid suffici histori larg practic cull minor outcom improv situat control design level histori captur suit note captur burden screen filter addit benefit small track histori user action easi unix complex graphic user interfac histori complex simul captur state process difficult cole jone record simpl click locat record interpret strongli prefer compon interact vacuum pump simul record click locat level record valv open pump start ad time stamp record state simul time pose design accommod effici histori record distinct drawn determinist stochast limit number user event system histori model compact design softwar engin concern gener applic scale larger simul histori data structur visual gener control panel search applic user learn benefit assum build learn histori benefit primari student review reconsid work product learn experi student send work peer mentor comment explor freeli advic instructor monitor student histori awar student problem learn modul defici expand specif ti specif applic vari student educ situat initi implement learn histori implement learn histori mechan context applic framework construct learn environ call simpl process learn rose goal develop learn histori mechan applic framework modul develop simpl gener learn histori anticip initi histori system provid instructor tutori built annot facilit commun instructor student record histori student simul run initi implement design goal mind figur oper simpl modul learn basic principl vacuum pump technolog semiconductor manufactur simpl modul develop simpl dynam simul visual repres realist behavior coupl guidanc materi softwar architectur enabl independ contribut develop repres educ content domain knowledg form simul model guidanc softwar develop user user interfac templat accompani softwar aid reduc softwar develop effort simul simpl modul primarili dealt process control simul involv sequenc oper continu determinist physic process main focu semiconductor manufactur applic includ simul river traffic flow major learn histori implement implement initi learn histori mechan simpl simul figur modul design teach basic principl vacuum pump technolog semiconductor manufactur simul emul configur evacu reaction oper pump valv input control pump valv output data consist total pressur measur pump chamber reaction chamber modul teach procedur produc extrem low pressur reaction sequenc action turn pump control vacuum channel open close valv learn histori domain time stamp activ action simul open turn annot comment audio reactiv messag statu report stabl pressur desir vacuum level model user modif simul paramet minim histori model determinist simul case simpl simul save reactiv event gener experi support gener movement time record simul gain greater histori model requir greater portion entir state simul perform challeng explor project simpli reproduc session modul reactiv suffici record state paramet time input control visual visual mean user interact histori simpl session show parallel control state pressur make appar consequ action chamber pressur bottom displai event chang state control pictur symbol green red pump open pump close type green line show period valv open pump green line valv close pump turn top displai figur show pressur valu point time visual displai make easi discern current state system period control correl control event effect chamber pressur featur visual annot mechan permit student instructor commun session note mark question displai icon case student add question instructor event occur instructor annot student histori question instructor build tutori demo instruct histori ad annot provid guidanc replai session histori replai save recal replai histori visual function histori icon repres simul event drag left event occur earlier replai annot edit reposit featur prove invalu tutori author fine tune tutori action tutori save oper thumbnail session replai displai correl output annot present text dialog simul dialog box dismiss subsequ implement supplement time run note drop timelin replai press button assum learner read add note examin run histori advanc learner instructor prepar demo tutori visual control time note visual place control speed option replai highlight element simul current control state histori flash attract figur visual annot user initi bottom half user top half show pressur valu histori produc simpl visual save email figur load replai replai produc sequenc annot flash control state make instructor simpli record sequenc review replai fine tune event sequenc edit add send class learner experi experi resum demo add question instructor learner start creat complet simul run record peer review test annot tool simpl time user feel confid power advanc user instructor design complex tutori adequ simpl default option burden novic control panel give user control level detail record rich histori essenti environ pilot studi order test earli design ran inform evalu user instructor run prepar tutori demo construct instructor oper control optim pressur level reach reaction chamber goal brought optim outcom subsequ instructor carefulli examin visual histori recogn kei event requir ad text annot present figur histori playback simpl modul refin classroom popul chose mix undergradu graduat student initi tester materi user learn histori knew vacuum pump make learner content histori tool round review sat user answer gener question record comment usabl problem user work short tutori vacuum pump prepar instructor includ demo session user email gener comment includ annot histori user discuss topic experi email end experi interview user independ result receiv email attach histori file particip email numer comment averag comment case attach histori annot particip comment specif simul run particip question answer return histori annot instructor particip found materi produc histori modul present demo notion neat tempt chang tutori materi arrang organ demo fine troubl run creat common request extend technolog freedom movement replai back back histori point nice fast forward jump point plai similar suggest made support initi simul set prior playback demo suggest chang set impress suggest chang set histori playback told support good idea complic figur edit histori file origin usabl factor mention clearer demarc live simul histori replai suggest obviou indic playback real estat put histori stuff main time time event demo futur direct studi conclud student found histori mechan valuabl learn eager experi power histori environ permit select replai interact deeper investig tradeoff dataset size capabl histori prove simul set provid ad function initi simpl visual histori system gener includ simpl framework requir develop set compon configur record simul histori set domain includ semiconductor river highwai traffic manag broad spectrum data control pattern applic area test gener complet compon encourag initi effort histori great potenti tool collabor learn hope futur work produc gener methodolog provid instructor learner capabl paper natur compon educ experi acknowledg author greatli assist julia harri steven friedman research prepar manuscript refer advantag disadvantag comput tool electr engin ieee tran educ learn assess journal tool structur reflect process proc frontier physic tutori physic proc frontier enhanc cooper teacher colleg press comput support learn complex problem commun acm user repeat action principl design histori proc human factor comput back altern behavior web browser back button annual human factor web held june collaboratori commun acm graphic web studi acm confer hypertext acm digit freedom privaci learn environ magnet reson support learn oper journal educ multimedia hypermedia macro proc user interfac softwar technolog histori edit graphic watch program mit gb learn journal water resourc system press educ semiconductor manufactur process dynam proc frontier jl tool build learn learn technolog workshop canada dynam interact simul reflect proc frontier ieee understand transport manag system perform learn maryland depart comput scienc technic report colleg md gw applic framework creat learn maryland depart comput scienc technic report colleg md design user strategi effect publ peopl revisit web empir find implic design histori intern journal academ press visual histori web inform retriev montreal tool inform human factor comput pittsburgh 
ML/549 6 bayesian explor mobil robot robert sim abstract work address problem robot explor task automat learn map environ mobil robot navig local explor mechan intend applic arbitrari independ world approach avoid arbitrari heurist preliminari result present discuss futur direct investig content introduct problem statement previou work approach comput geometri topolog autonom map construct problem theori bayesian analysi presenc uncertainti explor develop theori explor bayesian approach ad measur preliminari result simul instanti futur direct list figur polygon shortest rout spectrum environ mont carlo simul robot undergo uncertain motion simul result origin scene landmark laboratori scene probabl densiti function robot pose introduct order autonom robot oper requir knowledg task accomplish essenti aspect knowledg map inform enabl robot estim research devot signi amount work solv prior map tackl problem construct map attempt map acquisit problem fulli autonom set major commit modal assum world repres map de ne explor process discov aspect world reliabl model structur behaviour scienti explor real world termin practic robot explor acquir sucient inform perform task measur con denc work aim develop framework autonom explor accomplish task pose problem optim robot knowledg inform world prior work domain constrain sonar vision sensor restrict class model seek gener model facilit wider domain aim address open question pose approach explor principl contribut work theori explor accommod multipl correct world account uncertainti robot action balanc paper divid de ne account question sensor uncertainti problem statement review prior work explor problem present approach provid preliminari result futur direct work problem statement goal develop robust ecient method autonom explor unknown environ task odometri sensor read factor account plan move agent goal maxim certainti world problem compound requir higher accuraci part safeti limit choic model question comput tractabl light de nition requir criteria determin explor termin clear robust ecient solut requir care issu previou work section previou work relev explor problem approach comput geometri problem comput geometri provid insight relat task sonar map real world approxim term bodi research polygon immedi applic import question answer context cg complex cg problem dicult conclud relat problem hard world express term exact solut problem repres lower bound accomplish presenc provid measur evalu algorithm tackl uncertainti rst rout solut repres minimum distanc robot travel order discov repres polygon figur polygon shortest rout start point rout mark bold line rout problem relat art galleri problem chin de ne rout polygon close walk point visibl point walk start point polygon rout problem shortest rout start chin demonstr problem polygon hole problem simpler polygon chin develop algorithm orthogon polygon provid algorithm simpl polygon al develop algorithm solv problem ed explor singl face solv rout problem face partial knowledg class navig bug deal task nding path start point goal environ popul unknown obstacl explor strategi involv travel search obstacl obstacl encount taylor kriegman emploi build set local map base set visibl landmark topolog world consid section class cg problem relev geometr probe problem geometr probe consid problem determin geometr structur result measur devic probe context geometr probe relev result problem domain number discret site sampl order complet reconstruct environ probe model ect ideal behaviour absorpt broad survei wide rang probe model relat result signi interest nger probe introduc cole nger probe measur rst point intersect direct line object model perfect sonar measur sampl cole show probe sucient complet determin convex polygon probe model insu cient determin arbitrari polygon al modifi probe model develop cole probe follow arbitrari curv return point intersect polygon surfac normal point intersect probe author show probe sucient probe polygon edg relat interest problem geometr object veri cation set object target minimum set probe target discrimin object set detail survei result geometr test dudek develop collabor algorithm explor simpl polygon robot commenc explor posit adjac vertic polygon sweep free space travel edg maintain line sight entir polygon result descript topolog base planar decomposit polygon topolog autonom map construct earliest work aim address autonom explor problem kuiper byun work model world graph environ popul point line featur goal automat extract topolog vertic locat local maxima measur distinct subset sensori vertic graph recogn local signatur environ vertic connect arc de ne local control strategi travel vertex recognit veri ed procedur approach duplic gener voronoi graph deal issu de ning local control mechan move vertex vertex topolog approach relev work present al world repres topolog set visual problem local world recogn place base current input imag advantag emploi topolog dicult problem maintain robot pose absolut global refer frame introduc level abstract emploi infer understand command live room approach kuiper byun robot forc oper context extract adequ task requir special knowledg place graph dudek resolv issu de ning set metric island reliabl neighbourhood distinct place distinct place select measur combin inform quantit measur robot local select sensor task neighbourhood figur summar spectrum consid rang pure metric map pure topolog map requir metric inform dudek requir hierarchi gener ascend hierarchi problem theori bayesian analysi move problem rst examin theori motiv current approach task pure metric pure topolog figur spectrum environ commenc rst de ning pose robot repres global paramet captur state world de ned con gurat space de ne set admiss pose observ de ned output sensor belong space admiss sensor read repres raster set pixel intens valu imag acquir camera de ne map abstract method predict observ base pose robot map typic coupl set prior observ de work foundat problem theori problem theori concern problem infer base observ proce robot local imag robot encount posit express relationship local problem invert equat infer problem pose observ sensor imag problem theori rescu principl commit di erent pose imag repres pose robot probabl densiti function input imag map question comput analyt clear import paramet world light surfac geometri notion road map method predict upcom topograph map predict slope trail ahead util map local root remark abil rapidli search place predict match observ develop theori predict observ pose map embodi properti world enabl predict bayesian refer model pose repres subset paramet reader impli comput framework select refer refer member comput method comput explicitli explicit object regular observ spline interpol issu comput model address mackai ferri deal issu context object object repres set abstract infer simpl applic law reveal refer priori probabl densiti function robot prior refer evid map term assum uniformli distribut treat normal constant note prior evid uniformli distribut pose maxim likelihood observ solut equat pose robot underli probabl densiti function complic simpl method desir comput compact pdf varieti comput tool emploi repres probabl densiti function common tool mixtur model mont carlo simul mixtur model repres probabl densiti function nite sum elementari pdf gaussian repres weight term appli elementari pdf approach repres data mixtur model requir priori knowledg number mode data mont carlo simul comput approach probabl densiti function repres implicitli set outcom repeat stochast simul pdf pose robot repres set weight locat point con gurat space weight likelihood robot translat forward figur mont carlo simul robot undergo uncertain motion uncertainti robot head yield distinctli distribut figur courtesi ioanni locat point figur depict mont carlo simul result pose robot undergo uncertain forward motion note uncertainti robot head lead distinctli pose distribut number particl distribut particl repres pdf robot pose advantag mont carlo simul priori requir structur pdf disadvantag number particl requir adequ repres pdf exponenti paramet space faithfulli simul stochast physic system dicult problem mont carlo simul appli successfulli problem robot local visual arm relat turn attent task construct map environ note discuss problem data acquisit data arriv presenc uncertainti robot construct map execut action sensor read task involv measur uncertainti corrupt result uncertainti robot action compound success map eventu useless mention robot classic construct map pose problem probabilist framework problem repres world select probabl straightforward approach probabilist map construct emploi kalman filter updat robot posit posit landmark concept diculti pose kalman filter uncertainti embodi gaussian evolut actual uncertainti pose robot addit gaussian kalman filter assum problem linear extend kalman filter accommod formul rst linear neighbourhood estim taylor seri diculti approach linear lter diverg correct solut order avoid problem encount due assumpt linear gaussian turn gener framework pose suggest thrun al thrun work advantag markov assumpt law show sequenc action observ input probabl map maxim likelihood data method produc probabl map simultan probabl set robot base expect maxim iter strategi likelihood space oper expect maxim phase thrun em pose comput set robot action assum current estim map correct comput maximum likelihood map robot assum pose estim step correct process continu iter map converg local maximum recent thrun suggest increment method adjust map data arriv method exploit mont carlo pose robot distribut comput error robot pose backward time prior observ approach maintain set good start point gradient ascent likelihood comput forego cost search entir space map maximum likelihood estim em approach exploit context comput structur motion camera work signi exploit altern map emploi section note kalman filter approach motiv previou success robot local algorithm map construct present thrun satisfi speci underli uncertainti requir approach assum robot perform task explor collect observ build map make attempt instruct robot collect observ task section explor consid task local map turn problem explor major explor approach appli heurist increment discov unexplor space diculti method qualiti result imposs valid evalu problem address author context problem theori theori optim experi mackai seri paper address problem bayesian interpol function sampl observ presenc observ uncertainti mackai exploit shannon entropi show optim place obtain sampl predict observ observ place interpol function largest applic mackai work fundament theori deriv preserv deriv object function data select ferri emploi mackai framework work activ explor obtain object model show locat observ maxim predict base covari matrix set model paramet describ object formul process base local measur model direct agent sensor read climb simultan uncertainti gradient ferri exploit idea solv object recognit direct sensor place maxim expect reduct entropi probabl distribut object class case author bene fact construct object inform theoret approach appli problem robot navig roi thrun de ne object function navig goal posit world simultan aim minim entropi probabl densiti function de ning pose robot reach goal reason path result path plan method refer tendenc direct robot boundari certainti robot pose improv approach mackai reli import assumpt approxim assum pose robot indic deal uncertain outcom robot action assum distribut model paramet facilit analyt solut observ site appli gaussian assumpt order approxim local gradient uncertainti space make local move assumpt reason critic assumpt theori model phenomena give rise data assumpt agent direct repeatedli collect sampl site model lead improv denc model lead robot ignor region poorli repres mackai refer issu approach goal work develop framework accommod inabl complet captur data develop theori explor consid extens prior work topic explor environ proce develop theori motiv approach emploi approach account requir account uncertainti pose robot bayesian approach map sensor read equat comput probabl pose robot repres predict model theori measur likelihood observ base world context naiv approach data collect move place select action uncertainti pose robot maxim amount select pose maxim expect posteriori entropi log interest contrast idea navig approach roi move goal world minim entropi posteriori pose distribut maxim incorrect explor depend xed properti di erent place world alik matter observ uncertainti place versu ident place remain xed interest likelihood map current equat relat likelihood map face observ pose likelihood observ priori measur probabl map note independ absenc map inform theoret approach explor pose maxim expect reduct entropi probabl distribut state expect chang entropi term mackai de ne log dm repres prior arriv datum set collect data evalu equat requir integr space map pose diculti contrast method construct parametr model base data instanti map term explicit set paramet avoid instanti order paramet implicitli encod inaccess embodi sole set prior observ arriv continu investig insist map lead formul mackai simpli ed comput taylor seri map assum likelihood space distribut neighbourhood avenu explicit actual set observ yield observ space cross datum author investig equat simpli ed context work text mackai observ collect data point error bar interpol implic point abil map predict data weakest author unabl deriv explicit relationship chang entropi model predict observ nearest deriv suggest inform observ noisi appar contradict stem fact model predict noisi highli uncertain predict datum process forc draw line attack theoret unsound uncertain pose robot select pose navig select action execut robot maxim expect con gurat space assum moment con gurat space robot discret nite grid restrict action discret set move forward turn left formul task reinforc learn problem task comput polici select action order maxim expect accru robot princip diculti approach tractabl requir discret robot state straightforward question futur research ad measur direct robot place execut global optim data collect global optim action fact requir robot travel signi put risk due hazard fact physic cost execut optimum action robot lost travel goal accommod present object function repres cost execut action paramet determin desir rel weight expect chang entropi clear junctur sucient penal highli uncertain action question futur investig preliminari result simul section state work date rst instanti explor method simul robot environ figur simul result robot environ termin explor model environ occup grid point hold occ probabl posit intersect edg polygon occup grid initi indic state robot initi pose lower left corner map explor perform rst identifi set candid pose elig explor reachabl base current state occup occup grid simul pdf simul sonar scan posit pose sonar pdf highest entropi select optim pose scan figur show state occup grid iter algorithm model repeatedli select posit scan algorithm termin reason termin conclud section heurist incorrect mackai argument insuci model lead suboptim behaviour model sonar scanner heurist simul explor figur demonstr import issu face address signi interest pursu richer model turn attent task model probabl densiti function vision sensor real world set figur origin scene landmark instanti goal autom acquisit train input discuss section fundament mechan develop abil comput discuss prior work domain landmark learn introduc instanti appli bayesian context robot note prior work cite emploi bayesian formul present unpublish goal comput probabl densiti function camera imag pose map approach make simplifi gener entir imag concentr gener imag neighbourhood set fl visibl salient point probabl densiti function observ landmark distribut maximum likelihood observ comput explicitli function gener base learn mechan construct imag posit landmark set prior observ process scheme infer covari matrix distribut covari captur stochast natur sensor inadequaci model figur depict scene landmark learn uniform sampl pose space figur depict imag nearbi pose pose imag neighbourhood landmark reconstruct method comput formul mixtur gaussian landmark imag correspond gaussian probabl densiti function landmark pose learn predictor landmark point justi cation made choic mixtur model equat correct formul comput joint distribut consid possibl outlier landmark recognit result due limit machin summat model prevent catastroph cost higher uncertainti incorrect assumpt landmark observ disjoint figur depict pose space laboratori set depict figur mode distribut predict pose robot posit actual robot posit futur direct establish framework autonom inform discoveri context robot explor wide varieti issu requir discuss signi issu tractabl integr solv order comput entropi function requir simpli cation exploit mixtur model formul order deriv analyt solut lack explicit model handicap deriv address question model remain comput model present model model imag domain requir complic con gurat space work requir context evalu expect entropi uncertainti robot pose evalu set singl vantag point obtain sensor read question reinforc learn paradigm appli consid address infer geometr constraint context navig ad cost function aim reconstruct imag reconstruct sonar function equal interest issu futur work natur topolog environ express set landmark view di erent region figur laboratori scene probabl densiti function robot pose area depict correspond con gurat space robot darker region correspond imag space give rise di erent set visibl landmark opportun path plan infer base navig set visibl landmark approach li domain hybrid set topolog place local metric map signi gap domain robot research autonom explor domain machin learn studi make research consid task acquir data optim facilit task hand work aim ll gap theoret sound practic feasibl theori explor deriv inform theori speci robot direct acquir sensor read place abil model world weakest establish preliminari result valid part feasibl goal highlight import diculti overcom futur work continu seek marriag robust theori explor practic issu implement refer jd order induc set applic probe convex polygon ieee intern confer robot page ieee tal activ object recognit condit probabilist evid entropi map phd mcgill januari tal frank ferri viewpoint select navig entropi map seventh ieee intern confer comput ieee press wolfram dieter mark reid sebastian thrun collabor explor ieee intern confer robot san ieee press francesco frank ferri activ uncertainti reduc ambigu technic report mcgill intern journal comput vision chin shortest rout simpl polygon discret comput chin simeon optimum rout inform process joel sensor base part increment construct gener graph proc ieee confer robot page ieee press cole shape probe journal thrun algorithm mobil robot local ieee comput societi confer comput vision pattern recognit ieee june frank steven charl sebastian thrun structur motion correspond ieee comput societi confer comput vision pattern recognit ieee june rubin maximum likelihood incomplet data em algorithm journal royal statist societi seri duda hart pattern classi cation scene analysi york gregori dudek environ map multipl abstract level volum nov theori optim experi academ thrun activ markov local mobil robot robot autonom jose hugh simultan local map build natur featur environ sixth intern confer intellig autonom michael isard andrew blake condit densiti propag visual track intern journal comput allan jepson michael black mixtur model imag technic report project technic depart comput march kuiper byun robot explor map strategi base semant hierarchi spatial robot autonom benjamin kuiper byun qualit approach robot explor ieee workshop spatial reason page lo ieee john leonard hugh mobil robot local track geometr beacon ieee transact robot june john leonard han jacob feder ecient method concurr map local john dan robot ninth intern feng lu milio optim global pose estim consist sensor data registr proc ieee confer robot page ieee press vladimir sun kang dynam path plan terrain acquisit ieee transact robot aug mackai bayesian interpol neural mackai object function activ data select neural art galleri theorem algorithm oxford ioanni bug survei technic mcgill ioanni gregori milio explor unknown ecient reduc odometri error intern joint confer arti cial volum page august morgan kaufmann ioanni gregori milio collabor robust explor intern confer robot san april kathleen geometr probe test survei technic report rutger septemb nichola wolfram dieter sebastian thrun coastal navig robot motion uncertainti aaai fall plan nichola wolfram dieter sebastian thrun coastal navig mobil robot navig uncertainti dynam environ ieee confer robot ieee press recent result art galleri septemb ce shannon mathemat theori chapter mathemat theori commun illinoi shi carlo good featur track ieee confer comput vision pattern june ieee press sim mobil robot local learn landmark master mcgill juli sim dudek mobil robot local learn landmark confer intellig robot system octob ieee press sim dudek learn evalu visual featur pose estim seventh ieee intern confer comput sept ieee press steven interact reconstruct geometr probe septemb randal smith peter cheeseman estim spatial intern journal robot richard sutton andrew barto reinforc learn mit visual place recognit autonom robot ieee intern confer robot autom page ieee comput societi problem theori elsevi scienc publish taylor david kriegman explor strategi mobil robot ieee intern confer robot ieee press burgard probabilist approach concurr map local mobil robot machin autonom robot sebastian thrun bayesian landmark learn mobil robot local machin sebastian thrun learn map map indoor mobil robot navig arti cial page januari sebastian wolfram dieter fox algorithm mobil robot map applic map ieee intern confer robot san ieee press ferri autonom driven uncertainti confer comput vision pattern page lo june ieee comput societi press peter curiou machin autonom visual explor phd mcgill 
DB/1101 3 aachen technolog research group theoret comput scienc approach model queri video data jacqu extend version articl intern confer data rwth aachen informatik aachen germani abstract index video data essenti provid content base access technolog offer integr framework model queri video data concern video model found provid interest angl attack problem video applic system provid nice basi futur video system research provid solut video issu partial fragment video applic provid beauti challeng gener system provid support multimedia data data type requir techniqu manag solut paper develop data model queri languag video content base index retriev data model design object constraint paradigm video sequenc split set fragment fragment analyz extract inform interest put search find inform interest type inform entiti interest domain video video frame entiti repres data model fact object constraint present constraint queri languag infer relationship inform repres model languag clear declar oper semant work major consolid access queri constraint queri languag introduct recent progress compress comput store huge amount audio video media wide todai commun form home educ scholarli corpor effici comput exploit lack creat face develop advanc video virtual internet interact video conferenc video partial advanc applic integr video data complex manipul video analysi content retriev base semant requir research effort area comput imag data inform video data manag pose special challeng call techniqu easi develop applic facil user view video materi navig build sequenc facilit semant object featur video appropri index keyword free text describ semant object suffici addit techniqu state issu address video inform form facilit retriev organ inform effici present retriev video sequenc deriv adequ content descript satisfactori retriev condit end mandatori video data model power express sophist content proper usag queri video natur video import develop adequ data model queri languag featur system desir video secondari storag concurr support video inform share inform applic make analysi advantag gener video abil repres complex open extern world languag tradit system exist technolog design manag digit video class techniqu requir retriev automat process present visual content tool video exploit amount displai video model method develop specif queri languag concentr extract capabl queri video data mean set ad hoc incorpor investig theoret framework argu queri languag appropri design support video specif featur form sound basi queri languag point video data present interest challeng futur system cover rang task manag video content includ featur develop scheme oper data model express captur characterist inher video queri languag kind virtual edit perform exact partial fuzzi match consensu central role video plai research work find semant foundat repres queri video inform paper contribut direct framework present integr formal develop object sequenc paper build work hybrid data model video data constraint queri clear declar oper semant make develop simpl video data model basi object constraint paradigm object interest relationship object attach gener interv pair relat constraint queri languag infer relationship inform repres relationship object high level specif video data manipul model queri languag approach repres period time gener interv queri conveni declar natur previou research power constraint implicit specif tempor data gener interv set overlap fragment video sequenc model queri languag core video document archiv prototyp channel nation institut formal queri languag queri video data paper paper organ section discuss relat work section inform introduc index video sequenc section present definit section formal introduc video data model section describ underli queri languag section draw relat work advent multimedia comput standard power compress techniqu digit store common human video stream worldwid store minim function expect power aim content index retriev main approach fulli autom content index approach base interact fulli autom research system jacob weak content analysi focu specif exploit aid video content index system avi fascin issu context video data multimedia gener note present context imag video queri formul fall broadli textual visual system develop retriev visual data base imag relat bibliograph data focu textual languag work present closest complement retriev model imag base logic languag span main map content queri imag address dimens combin object attribut entiti mean relat object shape state declar equat case model tanaka data model focu capabl featur support schema evolut provid mechan share descript data video frame sequenc model object attribut attribut valu describ content semant scene sequenc video frame interv pair start frame end frame denot continu sequenc video frame introduc notion inherit base interv relationship mean notion share descript data successor interv merg overlap defin object object provid user queri languag retriev model descript definit relationship object content term semant object consid valu attribut adali al develop formal video data exploit spatial data structur store data kind inform object event role video divid sequenc frame constitut logic video set frame sequenc view frame segment event set attribut describ context event give parti attribut host valu philip attribut guest object event complex structur relationship object implicitli descript event develop simpl video queri languag retriev video interest extract relev segment video satisfi queri condit provid approach repres period time frame interest video sequenc gener video data model combin idea stratif segment approach object develop framework share video data model built model simpl video queri languag tempor interv oper provid work concentr structur part video order support video themat index base give textual descript content frame sequenc elabor structur descript content frame sequenc regard extend work descript content video sequenc mean class citizen object data model relat attribut explicit relat lead express relationship link object video frame call gener semant object interest gener model manipul level special spatial tempor express declar manner gener semant object relationship element make video sequenc applic model video overview inform introduc video index model discuss fundament aspect video present approach gener interv object discuss briefli relat convent approach video index mean defin inform carri part content video sequenc video data consist sequenc share attribut imag data posit rel layout video addit tempor relat attribut video data hierarch structur video index model provid facil effici captur addit attribut tempor natur inform abstract video content true time tempor manag video inform requir segment approach video index scheme sequenc split independ contigu time segment annot individu figur show basic segment broadcast easili timelin document partit individu handwritten descript public speak minist walk maneuv figur index segment segment light quick descript video document applic broadcast archiv approach critic davenport strict tempor partit result rough descript video document introduc call annot fact individu element interest singl tempor interv call shown figur stratif overlap user level descript figur show level decomposit document idea stratif interest fact descript extend stratif approach defin call gener interv contrast stratif time segment set time segment descript gener interv defin set interv provid time boundari descript handl singl object occurr entiti video document index annot period time object interest screen object report minist simpli associ object gener interv shown figur object interest applic depend tax armi educ polit financ broadcast cannon tank soldier talk public talk minist armi move figur index stratif gener interv trace presenc screen singl instanc refer occurr document gener interv occur report report minist time object interest figur gener interv index basic definit section preliminari concept design video data model underli constraint queri languag definit concret domain consist ffl domain ffl set predic symbol predic symbol ariti relat concret domain set integ comparison assum entail conjunct disjunct decid definit linear order inequ order inequ constraint formula form negat assum constant interpret countabl infinit set binari relat order interpret order concret domain complex constraint built primit constraint logic connect special denot entail write entail satisfi constraint unsatisfi techniqu check satisfi entail order constraint domain studi express power complex linear constraint queri definit domain constraint constant type set constant type denot set variabl rang finit set element type constraint restrict form set constraint involv set function note constraint deriv form rewritten fcg satisfact entail conjunct constraint solv polynomialtim quantifi elimin algorithm class constraint plai import role declar constrain queri answer definit interv consid order pair real number definit refer predic concret domain ir time interv repres conjunct primit linear order inequ constraint definit time gener time simpli gener set overlap interv gener time interv repres disjunct time interv basic video formal mathemat structur central notion video data model gener interv decid split video sequenc set gener interv analyz extract inform index basi content assum inform describ content interv length interv user basi index video materi establish link inform tempor locat video sequenc convent approach segment stratif tackl problem approach assign continu tempor interv video inform approach extend relat inform set tempor call gener interv semant scene sequenc continu video interv complex video handl main sourc inform ffl machin deriv detect color basic raw ffl applic specif desir video higher level semant data semant object refer semant video index natur video semant object index respect time video data model fact object defin video sequenc ffl set gener interv video sequenc split set arbitrari gener interv note gener interv overlap gener interv abstract object identifi ffl set entiti describ inform video sequenc applic inform interest exploit consid object determin analyz gener interv ffl set atom valu valu drawn concret ffl set relat relat relat object gener interv ffl set constraint describ time interv defin gener interv ffl map gener interv subset ffl map gener interv linear order constraint describ period time gener interv disjoint video data model ffl object object ident object entiti interest video sequenc refer object logic object syntact term queri languag logic oid uniqu identifi object word refer id logic level essenti type gener interv abstract object result split video sequenc set smaller semant object entiti interest video sequenc ffl attribut object attribut attribut defin object ffl relat argu time object model real world natur situat relat combin object lead natur relat encod natur handl relat prefer relat languag construct assum exist countabl infinit disjoint set atom ffl relat fr ffl attribut fa ffl constant fd ffl object ident oid id fid id distinguish object ident entiti object ident gener interv order associ time interv gener interv restrict form linear order inequ constraint valu attribut defin set element ffl primit constraint form ffl disjunct primit constraint definit set valu smallest set id fv definit video object consist pair ffl oid object identifi element ffl distinct attribut valu denot set attribut fa denot denot model framework recal concern movi alfr movi minut durat philip decid commit perfect crime prove privileg group peopl kill sake kill receiv punish kill friend david hide insid chest live room sign give parti invit friend david janet parent father mr david mr individu talk suspect david bodi room stand addit mentor intellig suspici prove worthi reput immedi understand circumst movi question gather clue find wrong gener correspond period time sequenc crime commit interv object chest object david call role philip david correspond period time sequenc parti interv object chest simpl extract content gener interv gi gi gi fo durat subject victim murder fo gi fo durat subject host fo guest fo role role role gi gi statement gener interv gi durat interv entiti interest fragment sequenc fragment sequenc deal murder attribut object object murder statement fact defin relationship object gener interv gi gi note tempor integ gener interv necessarili correspond singl continu sequenc video frame scene correspond singl continu sequenc frame describ period time gener interv disjunct atom constraint simplifi attribut durat gi gi mean constraint approach treat constraint organ class constraint attribut method attach addit inform featur video data model constraint queri languag present queri languag reason fact object video data model languag consist constraint languag top relat defin mean definit languag semant base notion extend activ domain extend domain gener interv object concaten languag interpret function symbol build gener interv concaten construct term form interpret concaten gener interv extend activ domain fix queri evalu gener interv object creat concaten object result concaten exist ad extend activ domain syntax manipul gener languag interpret function symbol construct complex term gener denot concaten languag term disjoint set constant symbol set union disjoint set atom set call object set gener interv object set variabl call object denot set variabl call gener interv denot denot gener interv gener interv construct interv construct interv term concaten oper defin structur result element defin structur ffl id id follow idea object id object gener function id id ffl ffl note mean obtain concaten result concaten lead termin execut construct rule definit construct formal definit constraint concaten gener interv definit defin predic ffl ariti predic symbol ariti ffl special unari predic symbol interv class gener interv object ffl special unari predic symbol object class object gener interv object definit predic symbol atom denot object object att att attribut constant negat call inequ atom definit rule languag constraint rule prefix constant symbol refer head rule refer bodi rule note restrict construct term head bodi rule construct term head call construct rule recal interest order arithmet constraint involv arithmet function restrict form set constraint involv set function definit rule variabl rule occur bodi liter variabl occur head occur bodi liter definit program collect rule definit queri refer queri tupl constant variabl give simpl exampl queri letter stand variabl letter stand constant queri object domain sequenc express constant output variabl gener attribut set semant object interest gener interv queri involv atom constraint comput answer set check satisfi constraint instanti queri gener interv object express queri object domain tempor frame express tempor variabl queri involv primit constraint complex arithmet constraint comput answer set check satisfi constraint queri gener interv object express fo queri pair gener object relat gener express queri gener interv object attribut express val infer relationship rule infer exist object defin relat hold gener interv object time interv overlap time interv express relat constraint entail defin relat pair gener interv common write object rule construct concaten gener interv common concaten fo fo semant languag declar semant semant recal denot set variabl call object denot set variabl call gener interv variabl var countabl function assign syntact express subset set variabl occur express syntact abbrevi ground atom atom ground rule rule definit program interpret consist ffl domain ffl map constant symbol element domain ffl map predic symbol relat definit valuat total function set element extend ident extend map free tupl tupl natur fashion extend constraint straightforward gener interv definit rule constraint valuat map variabl element rule true interpret valuat present ffl ffl present definit program interpret model rule valuat map variabl rule element definit mean program uniqu minim model theorem program interpret admit model includ admit minim model proof program interpret admit model set model show satisfi rule rule valuat base ext satisfi satisfi model satisfi rule instanc model minim model semant semant defin term consequ map interpret interpret interpret program subset ground atom formula built predic symbol languag element applic oper creat atom object construct show monoton continu comput iter fashion recal languag term countabl disjoint set atom valu set entiti set gener interv constant gener interv element defin definit set gener interv extens written ext set object ffl element ffl pair element element result concaten definit activ activ domain interpret note set element subset extend activ domain denot ext extens subset ext lemma interpret ext ext definit consequ program interpret ground atom consequ exist rule exist valuat base ext ffl ffl ffl satisfi definit oper program map interpret interpret fa consequ pg lemma oper interpret proof interpret show atom consequ consequ case ffl ffl exist rule valuat base ext satisfi lemma valuat base ext satisfi theorem oper interpret infinit proof atom show case ffl exist ffl exist rule valuat ext satisfi exist exist satisfi set variabl rule result appli variabl finit subset ext base ext satisfi ext lemma model iff proof interpret denot set ground fact consequ fa consequ pg element case ffl definit ffl exist rule valuat model satisfi interpret program rule valuat satisfi lemma model proof immedi lemma theorem program input minim model minim model coincid proof program denot minim model lemma model minim model minim model model minim model datalog set order queri shown evalu close form data complex rule languag arithmet order answer queri comput ptime data complex obtain lower bound complex queri evalu rule base queri languag futur work grow interest video theoret set understand relat model queri problem lead develop power system manag exploit video inform address problem develop video data model constraint queri languag definit retriev content video data primari motiv work object time interv relev video model absenc suitabl support structur tradit data model queri languag repres obstacl data model queri languag abstract visual video support model retriev semant data model style video independ content inform relat associ object video sequenc paper make contribut develop simpl video data model integr object constraint object maintain view inher video data attribut relat captur relationship object simplifi index video sequenc develop constraint queri languag reason object build sequenc function virtual edit applic languag declar natur express queri due complex natur video queri languag present facil user construct queri base previou queri properti inher imag data part video framework present natur appli imag data interest direct pursu ffl import direct activ research extend framework incorpor abstract mechan gener ffl import direct studi problem sequenc present exist research system approach provid automat sequenc capabl set sequenc templat predefin confin user explor sequenc order problem approach reli avail suitabl templat queri framework base declar graphic languag offer possibl flexibl specif sequenc present investig import research direct refer subrahmanian advanc video inform data structur queri process multimedia smith davenport stratif design environ random access video technic media movi context usenix usenix page dan video queri formul wayn niblack ramesh storag retriev imag video iii san page februari alexand aiken edward wimmer solv system set constraint ieee comput societi annual ieee symposium logic comput santa page timothi multimedia system version retriev wayn niblack ramesh storag retriev imag video iii san page februari alberto del icon environ imag queri ieee tran softwar alexand yoram queri constraint object acm sigmod intern confer manag data san page ns chang ks fu pictur queri languag pictori system novemb jan chomicki tomasz imielinski relat specif infinit queri answer acm transact june video retriev sequenc system acm transact inform system octob jacqu model queri video hybrid approach ieee workshop access imag video librari santa jacqu model queri video euromicro confer workshop multimedia david scott semiautomat video system wayn niblack ramesh storag retriev imag video iii san page februari myth semant video retriev acm comput decemb le video compress standard multimedia applic commun le mpeg compress review commun stefan grumbach su constraint symposium principl system san page june stefan christoph linear constraint queri languag express power complex daniel intern workshop logic comput complex page octob bernard paul victor jame norman iri hypermedia servic commun januari ramesh terri index video wayn niblack ramesh storag retriev imag video iii san page februari kato queri visual intern confer extend technolog page march rune roger model queri video data intern confer larg marku smolka definit relat constraint languag report ibm stuttgart octob john wei video manag system base advanc techniqu sethi ramesh storag retriev imag video iv san page februari joseph af high level queri languag pictori manag ieee tran softwar michael kifer jame wu logic logic program symposium principl system page march la jacob queri system video wendi mackai davenport virtual video edit interact multimedia applic commun juli marcu subrahmanian foundat multimedia system mecca anthoni bonner datalog transduc symposium principl system san page carlo logic reconstruct imag retriev sethi ramesh storag retriev imag video iv san page februari qbic queri imag content shape symposium electron san februari katsumi tanaka design implement system ieee transact knowledg data august jan peter letizia tanca declar graphic queri languag claud michael intern confer deduct volum page decemb peter revesz datalog queri set constraint georg gottlob mosh intern confer theori page januari lnc divesh raghu peter revesz constraint object intern workshop principl practic constraint program number page springer acm toi special issu video inform system acm transact inform system octob david toman point queri languag tempor symposium principl system page june video handl base structur inform hypermedia system intern confer multimedia inform system page mcgraw ichikawa video retriev motion journal visual languag zhang implement video present system sethi ramesh storag retriev imag video iv san page februari 
IR/2008 5 report inform servic research base mark octob swedish institut comput scienc human comput interact languag engin laboratori box sweden abstract present type web inform user inform broker collabor creat inform servic servic task inform maintain infer strategi user model inform broker obtain accur inform user adapt ensur user profil date consist user prefer illustr approach confer call collect distribut interest reader keyword adapt inform intellig inform user user profil introduct rapid develop inform sourc world wide web left reader acut problem inform overflow problem simpli inform retriev inform user requir aid summar retriev inform judg accuraci qualiti clear role human expert evalu inform inform broker servic exist todai util individu user profil tailor inform select individu user approach made system adapt automat user prefer servic person web radio station silver island immin problem inform broker situat select structur inform editor perspect typic inform broker servic util predefin classif schema inform reader individu servic select predefin categori approach individu reader select class inform orthogon real interest individu reader forc broker classif structur retriev inform imposs user type inform cover servic feedback inform broker obtain reader profil describ reader inform classif schema chang requir user chang profil approach basic idea exist parallel classif specif editor organ schema stabl time basic util intellig filter techniqu inform broker reader commun synchron classif schema editor support open adapt inform servic system adapt automat individu user make infer base type stereotyp knowledg user sourc knowledg human expert knowledg user equip system rule rule learnt system behavior group individu user repetit behavior sourc inform weak disadvantag knowledg reli advanc analysi user system user chang behavior problem appar domain inform rapidli chang highli unstructur analyz repres widespread user web filter inform adapt navig individu automat mean infer potenti deal problem approach infer user prefer base action singl user problem approach system deal inform problem address system reli group user model user recommend system servic firefli prefer individu user compar full user system suggest inform base fact user similar prefer pattern inform case prefer group peopl suffici satisfi user user interest singl expert regard import recommend larg group peer gener user judg relev piec inform base qualiti quantiti number peopl recommend inform broker role collabor filter main goal project util human resourc adapt inform system inform broker dedic expert collect potenti review literatur restrict area journalist produc articl specif reader group librarian organ incom inform direct reader inform process specif inform seek inform produc summari obtain inform common characterist role inform broker kind understand custom adapt inform broker collect inform evalu rel import includ inform disregard summar rewrit illustr differ origin sourc scenario inform broker benefit introduc group user model individu user adapt exist inform servic build user servic cnn custom swedish user explicitli set profil select set categori subcategori fit interest approach rise number categori fit user real interest prefer categor chang type inform user chang profil adher categor user seldom chang profil set profil reflect user true interest individu user adapt deal provid categor automat maintain learn user action system order introduc individu user addit task inform maintain rule stereotyp user structur inform rule applic essenti sourc inform task term profil set inform obtain inform immin inform accept addit task definit requir mainten inform structur task made simpl assum inform broker interest detail algorithm user model architectur implement system implement idea system support filter confer workshop call particip user individu review call set remind deadlin avoid uninterest user set filter retriev person select call organ person manner filter maintain semiautomat mean servic access web remind receiv email sm version servic implement evalu short messag servic messag servic provid european gsm mobil phone network technic detail servic built agent architectur consist number special agent person servic assist agent handl interact user central point interact user agent architectur user profil agent agent store prefer user user profil base inform user action receiv agent user inspect chang profil agent agent filter call user remind agent agent remind servic user agent user interest notif prior deadlin submiss paper remind agent handl send email sm user time agent agent handl transact store confer call call enter editor architectur prototyp version editor explicitli repres architectur call classifi enter tool editor part user collect agent forthcom log agent agent access agent enabl agent record event agent commun kqml messag content repres ground prolog term ontolog user commun agent user interfac remind agent user interfac user commun user profil agent agent share applet figur agent architectur design open extend kind inform servic discuss aspect agent agent figur denot agent consist half devot user interact softwar agent architectur agent span activ requir interact format access type agent remind agent complet overlap agent user set remind agent perform function set remind confer user control role reserv user user profil agent maintain user profil base inform user action receiv agent user mean inspect control user decid activ deactiv infer profil screen shot present call addit user profil function call part search delet put remind call figur show screen shot summari call present profil concern design user profil simpl maintain achiev requir user editor bound predefin ontolog formul filter structur call achiev profil annot simpl work filter mechan user inform broker essenti free defin classif enabl inform schema partial synchron filter mechan version reli type annot chosen buzzword confer call simpli set term chosen editor confer call user set profil term buzzword interest note intend buzzword keyword keyword typic aim describ topic style confer categor schema aim reflect subject interest editor user pick stabl categor predefin expect qualiti reflect trend research commun restrict describ topic includ peopl program place buzzword commun channel user inform broker user profil handl figur show screen shot user profil mainten screen left part activ profil shown list posit neg buzzword user delet buzzword add buzzword profil user tool aid ad buzzword select buzzword menu buzzword servic menu activ select buzzword candid profil activ type arbitrari buzzword candid user accept system suggest profil replac profil system suggest screen shot user profil manag broker profil handl inform broker maintain list buzzword servic list automat updat buzzword introduc system current buzzword introduc editor ad list version includ buzzword user ad profil broker free inspect modifi list ad buzzword show modifi exact formul delet buzzword task maintain list buzzword simpl variant editor task maintain rule stereotyp buzzword determin system learn user prefer buzzword list annot incom call broker inform retriev tool simpl extract subset buzzword found call text broker inspect result add delet buzzword call broker enter list buzzword maintain annot call expect happen lot confer call limit frequent set buzzword reach stabil local extern usag profil user profil wai system extern determin call retriev agent agent retriev call match buzzword call order local agent fit user interest call match keyword list lower number match buzzword shown figur user profil buzzword term make user interest call posit keyword fit local sort doubl usag user profil implement step split user profil local part extern part split import privaci perform reason local user model user model candid candid actual profil candid profil actual profil shown figur set candid buzzword intern system shown user set candid buzzword candid profil maintain local profil infer implicit inform user prefer user action action user make call indic interest system make user set call set candid buzzword candid profil maintain simpl learn algorithm user perform action system kei user buzzword call user act ad set action remind buzzword slightli posit small neg keyword occur call user act depend type action buzzword reflect consist user dislik call annot buzzword occur save delet approach algorithm ensur buzzword buzzword rare approach keyword ad candid profil approach keyword eventu delet set candid buzzword candid profil actual profil base predict user action user make action delet system map buzzword call fit profil call high rate system predict action save remind action rate system predict delet user actual action predict action system overlap user action support profil support profil gener simpli calcul number user action profil predict chang differ support candid actual profil determin system present activ suggest user chang profil simpl structur local user thing adjust make work practic expect tune stabil profil buzzword consid good ad candid delet actual profil candid profil motiv activ suggest hope tune stabil system compar log actual user interact system studi central featur filter sort simplist list buzzword simplic structur burden creat maintain profil annot user inform broker form major risk design user editor commun suffici list buzzword studi perform design specif aim find buzzword structur suffici user inform broker commun studi gather inform number deal potenti problem commun editor user addit trial collect log data tune user model algorithm evalu usabl local user model perform algorithm tune evalu buzzword analyz log data set user interact system expert inform broker inspect user profil util inform annot call seek answer question reader chang profil set reader type menu suggest user find cluster user profil similar interview user subject experi system interview direct find user higher express power formul user find review user make possibl categor call person categor call suppli broker studi start begin august result end year futur extens main studi establish list possibl fuse role inform broker reader obviou benefit scenario add confer read simpl extend annot call ad access restrict make modif call interest extens line make util distribut approach tradit architectur aim move servic connect client call difficulti ensur user obtain duplic call call extens usag text summar inform retriev elabor extens support collabor filter reader benefit review present typic profil group user editor recommend buzzword base peopl annot call read user sign call enter editor arbitrari call fit profil inform structur concern list retriev call formul profil possibl user sort incom call individu categori set singl user option set profil select call sort categori inform broker suggest categori flavor possibl chang time express power profil ad type search buzzword limit type inform locat program futur work servic show edit adapt inform servic larg complic task inform broker edit adapt ad individu user inform obtain mean organ inform suit reader buzzword function requir studi wai enhanc term user editor hint buzzword term structur buzzword annot kei issu find find solut strike balanc express editor user commun effici acknowledg project fund swedish research institut inform technolog swedish board technic develop refer cabl special issu recommend system vol march justifi user explain user mitr corp kqml agent commun languag bradshaw mit cambridg june edit adapt combin human machin intellig achiev filter inform stock comput report australia agent reduc work inform overload vol pp acm press social inform algorithm autom acm june servic contract negoti support open servic environ australia interact ab 
IR/2343 5 intern symposium intellig data analysi activ hidden markov model inform extract tobia er christian stefan wrobel po box germani germani abstract inform extract html document requir classi er capabl assign semant label word word sequenc extract complet label document markov model techniqu learn classi er challeng task learn hidden markov model partial label document train rst give detail account task loss show minim hmm describ em style algorithm learn hmm partial label data present activ learn algorithm select unlabel token user label studi empir activ learn reduc requir data label qualiti learn model achiev amount user ort introduct enorm amount inform unstructur textual tool inform extract enorm import tool identifi relev inform document convert structur format xml document rst algorithm set rule research turn learn extract rule document approach fail provid robust inher variabl document led recent interest hidden markov model markov model algorithm tag hidden markov model inform extract requir train document label token manual semant label process concentr task learn inform extract model partial label develop hmm learn algorithm develop activ hidden markov model select unlabel token document user label idea activ learn algorithm identifi unlabel observ label user algorithm classi cation cluster regress present rst algorithm activ learn hidden markov model paper organ give formal account task loss function section descript minim hmm present detail em style algorithm learn hmm partial label data extend algorithm perform activ learn section report experi activ hidden markov model section inform extract problem hmm begin give de nition task consid paper document sequenc observ correspond token document token vector attribut gener collect nlp tool attribut includ word part html properti paragraph task attach semant tag token observ left untag tag extract algorithm map observ sequenc singl sequenc tag fx xn assign handl problem de ned joint distribut document tag sequenc de ne error rate extract algorithm problem de nition assum document nite token de ne error probabl token tag write ith tag return token state task learn model partial label document de nition distribut fx xn observ space model set vector drawn task model minim error hidden markov model robust statist method structur analysi tempor data hmm consist nite state fs sn probabl probabl start state ij jq probabl transit state state probabl distribut jq observ inform extract observ token tag correspond target state sn hmm background token tag emit hmm state sn target state convert html phone directori hmm node observ sequenc extens correspond state sequenc task address rst knew hmm assum gener observ section discuss learn hmm partial label observ sequenc tag sequenc return minim observ return tag maxim probabl mean identifi sequenc state maxim return tag correspond state token brie describ element hmm algorithm refer detail discuss forward quanti es probabl reach state time observ initi part observ sequenc jq backward variabl quanti es chanc observ rest sequenc state time comput step procedur step calcul probabl observ sequenc model express probabl state time observ sequenc term step tabl observ sequenc gener hmm minim error return sequenc state maxim tabl show assign semant tag token document error minim learn hmm partial label document hmm inform studi learn paramet hmm set fo sequenc assum user label token tag express mean label function fo fx xn tabl algorithm input observ sequenc ot hmm ij ij tabl inform extract hidden markov model input document doc wt hmm set tag xn target hmm state sn call po gener sequenc ot augment vector word attribut attribut refer properti word paragraph occur call max output iw output user label token mean hmm state emit token label possess tag impli hmm background left unlabel state algorithm maximum likelihood hypothesi maxim obviou relat ml hypothesi hypothesi minim expect error sampl ml hypothesi prior probabl account regular mechan appli order avoid nding ml explain data priori laplac smooth addit maximum entropi suitabl regular instanti em algorithm main diculti algorithm address order estim transit emiss state sequenc correspond observ sequenc observ label transit emiss probabl order calcul probabl state observ sequenc algorithm start random interleav state probabl estim transit emiss probabl base calcul state probabl bootstrap procedur converg stabl optimum small error rate elabor element algorithm requir explain modi refer reader detail explan variabl de ne modifi de nition introduc adapt algorithm partial label sequenc probabl transit time de ned equat calcul equat equat split hand side equat equal de nition factor de nition introduc term ned equat residu equat jq ij ij jq ij case token label forc tabl label indic hmm target state set probabl renorm step tabl show result algorithm maxim likelihood token sequenc obei constraint manual ad label theorem document complet unlabel unknown iter stai constant case reach maximum document complet label algorithm stabil rst iter maxim tm likelihood label activ hidden markov model unlabel document obtain easili inform extract label token document ort activ learn tabl algorithm partial label observ sequenc input set token sequenc tm label function fo xn number state option initi paramet set initi model provid initi make ij probabl bg repeat call procedur calcul ts unknown ii iii iv unknown ts ij ts ij ik vi ts vii ts transit time frequenc start state ij ts ts number transit number transit observ valu ts ts frequenc observ state return probabl ij stai constant iter output paramet approach util amount user ort ectiv unlabel label interest object minim extract algorithm label token tag maxim jo deviat optim strategi paramet estim di er true paramet state state max di erenc probabl state con denc state con denc valu margin de ned empir instanc low margin relev adjust hypothesi paramet de ne margin token read time di erenc highest highest probabl state max fp max fp max max margin quantifi activ hmm learn algorithm rst learn initi model set partial label document determin margin token start user label token low margin algorithm previou paramet initi model adapt data experi gener hmm variabl number background target state random hmm gener unlabel observ label number initi token drawn random studi error develop number addit label ad observ sequenc strategi label randomli drawn strategi label token smallest correspond activ hidden markov model control strategi select token largest margin expect perform di erent hmm size hmm consist background target state state emit observ randomli drawn probabl gener sequenc initi unlabel observ hmm possess tabl activ hidden markov model input set sequenc tm label function sn number state queri paramet nq call algorithm determin initi paramet repeat call algorithm ts jo max max user label nq unlabel token smallest margin jo updat label function estim model paramet algorithm initi model user tire label data return hmm paramet hmm consist state curv figur averag lean problem initi sampl unlabel token medium size hard learn label token drawn random figur slight signi advantag random token select select token small margin dicult token begin bene cial averag phase token select small margin gain small advantag bene margin strategi initi sampl label token drawn token smallest margin select small hmm learn problem unlabel data rel problem complex bottom line error reach label margin strategi label token random strategi activ learn small margin exampl initi random token bene cial case base level error reach exampl activ exampl regular learn ve time fewer label exampl bad strategi case experi show class hmm dicult token learn begin result higher error rate train start label randomli drawn token activ hmm dicult token initi signi improv achiev regular hmm result time fewer label exampl label token random margin larg margin label token random margin larg margin label token random margin larg margin label token random margin larg margin label token random margin larg margin label token random margin larg margin label token random margin larg margin label token random margin larg margin label token random margin larg margin fig error rate activ regular learn number label token easi initi sampl label token drawn random medium size initi sampl initi label larg initi label discuss relat work de ned hidden markov model oper partial label observ sequenc de ned margin token measur diculti experi show import label dicult exampl low margin observ activ hmm requir time fewer exampl regular hmm achiev high level accuraci altern hidden markov model inform extract usual distribut ij altern model condit probabl jq rst consequ number probabl estim jqj jqj jqj model comput altern model render imposs hmm text classi cation model algorithm requir token label state modi cation appli altern model detail comparison model restrict model discuss token sequenc nest tag token label tag correspond cascad markov model solv problem gener case partial label hierarch hidden markov model adapt partial label token analog adapt standard hmm refer semant web road map intern world wide web brant cascad markov model ninth confer european chapter associ comput jordan activ learn statist model journal arti cial intellig mark dan dayn andrew tom kamal slatteri learn construct knowledg base world wide web arti cial inform extract world wide survei technic report norwegian comput tishbi hierarch hidden markov analysi applic machin ralph grishman beth sundheim messag understand confer histori intern confer comput thoma hofmann joachim buhmann activ data cluster advanc neural inform process volum hsu dung gener transduc semistructur data extract web journal inform special issu semistructur ander krogh neural network cross activ learn advanc neural inform process volum page kushmerick wrapper ecienc express arti cial andrew dayn fernando pereira maximum entropi markov model inform extract segment seventeenth intern confer machin rabin tutori hidden markov model select applic speech recognit wrobel learn hidden markov model inform extract activ partial label text technic andrew rosenfeld learn hidden markov model structur inform extract workshop machin learn inform vapnik statist learn theori 
DB/887 3 analysi rule xml jame bailei peter wood depart comput melbourn school comput scienc inform london abstract xml domin standard store exchang inform area data wareh rapidli grow technolog support reactiv function xml repositori rule automat perform action event natur facil support function studi eca rule context xml data defin simpl languag eca rule xml repositori languag illustr mean syntax semant formal investig method eca task ad complex xml set compar convent activ introduct rule provid reactiv function includ activ workflow network technolog implement busi process paper simpl eca rule languag provid reactiv function corpora xml document motiv work xml mechan data wareh recent extens commerci system support view export xml discuss increment mainten view xml data valid input data maintain audit trail data analog convent eca rule integr technolog provid kind reactiv function corpora xml document type eca rule mechan automat broadcast inform subscrib content relev document chang emploi flexibl mean maintain statist document web site usag behaviour kei recur theme success eca rule techniqu tool behaviour part paper explor techniqu eca rule languag xml closest work describ languag interact xml sourc web incorpor eca rule detail syntax semant rule analysi address discuss extend xml repositori eca rule order support activ extens xslt lorel languag handl updat event xml document insert delet event section analysi techniqu develop convent activ appli extend languag detail discuss specif applic approach push technolog rule action method updat trigger rule recent paper defin activ rule languag concern rule analysi rule syntax describ similar defin rule format base definit trigger rule execut semant model gener insert delet xml data bulk involv document fragment unbound size describ semant updat sequenc smaller updat depend content fragment trigger execut interleav execut smaller updat treat updat atom trigger execut invok complet toplevel updat semant produc result updat question futur research determin respect suitabl applic relat work discuss monitor subscript xml support subscript web document set alert monitor simpl web document monitor queri processor perform complex event detect send notif event trigger engin perform includ creat version xml document focu reactiv function highli tune specif applic extens xqueri languag incorpor updat oper refer reader paper review provis updat facil xml manipul updat oper express action support eca rule languag includ renam replac specif updat multipl level document trigger discuss implement mechan delet oper underli relat store xml provis eca rule xml level consid paper structur section defin simpl languag eca rule xml document languag illustr mean syntax semant formal section discuss techniqu rule includ issu trigger activ relationship sophist analysi techniqu base abstract interpret section discuss techniqu execut eca rule section conclud remark outlin direct work eca rule languag xml consist set xml document eca rule xml form event condit action illustr eca rule mean exampl defin syntax semant formal xml consist document list book magazin held element isbn number book stock greater eca rule updat inform document insert book insert insert true variabl success bound newli insert book node detect event part rule condit part rule check book greater action part rule insert copi isbn node child phrase true isbn node ad exist children eca rule similarli updat inform delet book delet true delet isbn node child match isbn delet book delet xml consist document store amount sale product store document hold inform includ store sold amount eca rule updat document product amount ad insert true delet insert true action rule delet store parent insert product store exist child product action add store child product net effect copi store remain product note attribut fragment xpath paper includ function attribut valu type idref xpath graph structur document attribut type id idref discuss view updat data xml consist restaur inform includ ingredi document view deriv list cafe ingredi ingredi element mushroom ad cafe eca rule perform view mainten insert insert true result document introduc eca rule languag inform syntax semant eca rule syntax event part eca rule express form insert delet xpath express evalu set node rule trigger set node includ node case delet case delet variabl condit action part eca rule set instanti set delet node return condit part eca rule constant true conjunct xpath express evalu xml document chang event form rule event part condit refer variabl evalu instanti document evalu document action part eca rule sequenc action action execut xml document chang event form rule event part rule condit queri evalu true call set document eca rule set candid document eca rule fire set candid document action express form insert delet xqueri express xpath constant true xpath qualifi insert express specifi set immedi insert express refer variabl construct instanti rule condit queri evalu true refer singl construct option xpath qualifi evalu child node insert form insert sibl insert form insert sibl true order insert delet express specifi set node delet refer variabl order appli analysi techniqu discuss section assum xpath xqueri express eca rule restrict simpl defin extend techniqu larger fragment xpath xqueri languag area work xpath fragment disallow number featur full xpath notabl axi ax function syntax simpl xpath express denot string denot element attribut observ phrase updat happen placement restrict occur root node document underlin term refer paper xpath express start establish call document function function path express refer variabl variabl express xpath express call qualifi delet qualifi xpath left path node call path distinguish path express node end distinguish path distinguish leaf express result xpath express set match distinguish leaf express result type denot element denot element result type determin distinguish path leaf string leaf leaf infer replac occurr left start substitut distinguish path xpath express occur event part rule step element replac separ occurr replac step step separ delet step separ term path result modifi distinguish determin leaf modifi distinguish path condit eca rule result type conjunct ingredi event part rule insert result type conjunct determin distinguish path substitut replac delet replac final delet leav modifi distinguish path xqueri fragment disallow full express keyword essenti permit part express syntax simpl xqueri express modifi distinguish path simpli determin result origin path xqueri express xpath express element constructor element constructor empti element element sequenc element content element list attribut attribut list empti equat attribut attribut list attribut string express element content element constructor express express xpath express brace brace evalu result insert posit element constructor attribut result type xqueri express denot node type element type suffix node root arbitrari captur fact result xpath express return set node root tree determin xpath express singl node type element constructor form tree element express ignor string express determin result type delet replac modifi tree tree repres xpath express xqueri express insert action rule store type xpath express xqueri express assum result type repres xpath express eca rule execut semant rule execut semant procedur pseudocod input execut xml consist set xml document schedul consist list updat execut updat pair action action part rule set pair delta identifi document appli delta set instanti variabl gener event condit part rule respect document time rule event condit part call set instanti set delta rule identifi number order prioriti common standard trigger assum rule loss assum schedul initi eca rule execut consist action initi set document delta action initi schedul singleton form head tail function head return element list function tail return list minu element function execut action head schedul set document refer updat perform candid document refer updat perform instanti candid document return pair db result updat delta arrai set delta event part rule respect document function determin set candid document rule respect updat execut set instanti variabl candid document evalu rule condit document case variabl occur condit rule case condit evalu element subset condit evalu true nonempti set determin return candid document pair set instanti wrt return candid document occurr condit rule case condit evalu document evalu pair set set instanti wrt return candid document number action action part rule jth action rule statement ensur action rule order schedul action pair set consist set candid document rule set delta document rule consid order prioriti statement action rule fire schedul front action rule execut proce fashion schedul empti rule execut possibl rule analysi techniqu import develop set rule discuss techniqu section observ number observ make execut ffl gener undertak simultan set updat insert result order set insert common parent construct determin order respect exist area work extend eca languag captur order relationship ffl coupl rule coupl mode assum mean action rule fire result current updat head current schedul coupl mode eca system defer detach examin applic impact context area work ffl bind assum instanti variabl determin time event part rule detect bind possibl examin implic context area work coupl bind choic made correspond standard trigger ffl eca languag encompass analog trigger relat activ occurr variabl rule action execut document time rule fire trigger occurr action action execut instanti document trigger eca rule behaviour analysi eca rule activ number approach literatur context relat kei analysi question termin rule set eca rule termin initi event initi rule execut termin trigger activ relat rule determin set eca rule rule trigger rule action gener event trigger trigger graph repres rule direct arc vertex vertex trigger acycl trigger graph impli definit termin rule execut activ graph repres rule vertic case arc distinct vertic condit chang true execut arc vertex condit true execut action acycl graph impli definit termin rule execut trigger activ graph combin method call rule reduct result previou method vertex incom trigger activ arc remov outgo arc remov vertic repeat vertic procedur result vertic rule set termin investig abstract interpret termin eca rule eca rule abstract abstract repres number real real schedul abstract interpret approach costli techniqu approach focu trigger activ relationship pair rule approach consid global relationship relationship evolv rule execut natur question analysi techniqu develop eca rule relat xml set translat set xml document eca rule relat form appli previou analysi techniqu rule problem approach result loss trigger graph transform translat xml document number unari binari relat relat state element relat semant element identifi element identifi relat semant data node identifi relat child store xml document content insert delet element eca rule form child condit action trigger graph rule relat modif trigger suffici translat xml data eca rule relat form techniqu develop conduct analysi directli discuss techniqu trigger graph xml eca rule order determin trigger relationship xml eca determin action rule trigger event part rule insert action trigger insert delet action trigger delet event insert action form insert rule insert event ev form insert rule ev independ recal section denot result type tree node element leaf node possibl suffix indic node root arbitrari set path set xpath queri fe event ev independ action queri independ queri set xml document queri option part insert trigger develop safe test base distinguish path recal section construct modifi distinguish xpath express path occurr element attribut occurr oper attribut occur end attribut children ev event action set element singl set attribut interpret singl occur xml modifi distinguish path interpret regular express alphabet overload regular express denot altern symbol alphabet regular express denot altern symbol alphabet form replac occurr occurr occurr occurr distinguish path match path element termin attribut document xml db string languag denot xpath express set action rule xpath express event ev rule modifi distinguish path impli xml document db test empti intersect regular languag standard techniqu intersect conclud independ rule trigger rule arc insert trigger graph test consid distinguish path conserv test case trigger state event insert action insert set queri deriv action singl queri ident distinguish path xpath express event method conclud rule action trigger rule event ev exact structur insert action element children action trigger ev requir child test detect develop test area research test delet action trigger delet delet action form delet belong rule delet event ev form delet belong rule view xqueri express deriv queri set insert case simpli xpath singleton set trigger independ trigger graph set eca rule acycl graph impli definit termin rule activ graph xml eca rule order determin activ relationship eca determin action rule chang condit part rule case activ arc action rule leav condit part activ arc tabl illustr transit condit absenc negat rule condit column show truth subsequ column truth insert delet ni rue rue rue rue case clear tabl absenc negat rule activ action insert condit condit xpath express requir return result distinguish path plai role xpath qualifi interest exist path document match distinguish path insert element occur qualifi turn true insert element turn condit true qualifi distinguish path similar test condit analog introduc rewrit condit condit condit turn true element ad element element element ad element element child procedur determin insert involv construct set xpath express qualifi distinguish path object condit chang true result insert condit chang true result insert start set fc set conjunct proce conjunct number condit ad singl step decomposit element ffl condit form replac ffl form replac delet step form replac occurr ensur occur end form replac uw form delet add condit case depend structur qualifi ffl match nontermin grammar simpl xpath add ffl match nontermin add ffl match add ffl match add ffl match match nontermin add match nontermin add decomposit process continu condit action rule insert section construct queri set test regular express intersect order check independ condit chang truth true deem insert activ arc conserv area research develop test condit eca rule set condit construct decomposit procedur singl condit modifi distinguish path substitut insert action activ ffl insert ffl insert book ffl insert bookstor book child child set queri gener insert action queri interpret regular equal interpret regular express mean condit insert possibl independ activ arc ad rule insert action rule condit condit eca rule substitut drop set initi condit condit condit interpret state insert chang condit true ingredi insert child insert child insert child restaur test detect possibl correctli insert activ arc case rule activ leav condit true analysi case rule obtain develop notion analog properti eca rule relat set rule execut action make condit absenc negat rule possibl rule action delet subsum conjunct condit delet action delet queri test simpl xpath express eca rule constrain provid oper qualifi decid test determin rule decid larger fragment xpath languag open problem fragment xpath properti develop conserv area research activ graph set eca rule acycl graph impli definit termin rule trigger activ graph combin rule reduct method emploi abstract interpret rule execut abstract interpret approach eca rule analysi discuss appli present xml set approach techniqu focu trigger activ relationship pair rule approach natur pair rule insert delet rule delet true insert easi trigger activ graph rule solid arc trigger broken arc activ trigger activ rule reduct method rule set abstract interpret approach rule termin test consist run abstract rule execut initi singleton abstract initi abstract condit unknown rue invoc abstract execut definit termin set eca rule conclud set rule deem possibl initi abstract schedul consist action rule trace abstract execut rule denot condit queri rule action rule iter schedul unknown rue rue rue rule iter rule execut rule iter remain make true conjunct prior execut made conjunct iter falsiti mean rule fire point conclud rule rule trigger rule execut termin iter initi abstract schedul consist trace abstract execut rule iter schedul unknown rue unknown rue rue rue execut iter iter rule fire conclud rule rule trigger rule execut termin iter conclud termin rule set eca rule execut number techniqu execut xml eca trigger relationship rule deriv techniqu discuss section order event detect function section action appli trigger rule evalu rule event queri set set empti document dynam maintain activ relationship rule rule section avoid evalu condit queri true modifi function section individu express condit action part rule recent result express document satisfi constraint conform schema document type definit rule condit action abstract interpret predict sequenc rule action execut current produc version rule execut semant sequenc techniqu convent activ appli equal xml set techniqu opportun rule execut specif sequenc rule abstract common set queri evalu techniqu rule compil rule execut simpl eca rule languag provid reactiv function xml develop techniqu trigger activ depend xml eca rule combin techniqu recent theoret result analysi approach eca rule develop convent number techniqu behaviour xml eca rule time techniqu eca rule xml data determin trigger activ relationship eca rule complex xml set relat determin effect rule action simpli matter match updat relat potenti event bodi rule condit associ implicit semant comparison set path express requir imagin xslt transform sourc document kind view document exampl paper xslt process entir sourc document updat order produc document detect updat finer granular activ rule updat document xslt requir result tree gener appli transform sourc document work open direct ffl explor practic effect test independ queri trigger analysi activ investig possibl develop ffl explor decid independ larger fragment develop conserv approxim case properti ffl extend express part eca support negat rule ffl incorpor inform document type order analysi inform possibl rule ffl investig coupl mode bind mode eca ffl develop analysi techniqu properti eca rule confluenc rule execut rule prioriti permit refer jl wiener increment mainten materi view semistructur data proc int conf larg page bs yesha relat transduc electron commerc editor abstract interpret declar languag elli push technolog person event correl proc int conf larg page hellerstein behaviour product confluenc observ determin proc acm sigmod int conf manag page acm hellerstein static analysi techniqu predict behavior activ rule acm lv srivastava minim tree pattern queri proc acm sigmod int conf manag page bailei abstract interpret framework termin analysi activ rule proc int workshop program lnc page activ rule partial evalu abstract interpret proc int workshop program dynam approach termin analysi activ rule proc int conf comput logic lnc paraboschi improv rule analysi mean trigger activ graph rule lnc page paraboschi runtim analysi activ behavior ieee transact knowledg data barali widom algebra approach rule analysi expert system intern confer larg page barali widom algebra approach static analysi activ rule acm ceri activ xqueri proc icd paraboschi activ rule paradigm vldb paraboschi push reactiv servic xml repositori activ rule proc widom practic applic trigger success issu proc int conf larg page ceri fraternali design applic object idea methodolog paraboschi web site gener applic proc int conf larg page view larg scale xml repositori proc int conf larg page deutsch tannen integr constraint xpath fragment proc int workshop knowledg meet mf suciu effici evalu xml queri proc acm sigmod int conf manag page ay suciu queri conjunct queri regular express proc acm symp principl page acm hopcroft ullman introduct automata comput ishikawa activ distribut system proc web dynam simon activ commerci experi activ rule page cochran activ featur activ rule page monitor xml data web proc acm sigmod int conf manag page hr principl program analysi editor activ rule system shasha effici match system proc int conf cooper inform system page zg ay ds weld updat xml proc acm sigmod int conf manag page wadler formal semant pattern xslt markup technolog widom ceri activ system san pt wood xml pattern proc int conf comput logic lnc page pt wood simpl xpath express proc webdb fourth int workshop web page world wide web consortium xml path languag version novemb recommend world wide web consortium xsl transform version novemb recommend world wide web consortium xqueri xml queri languag june work draft 
DB/124 3 monitor network log activ lane sara daniel lanc david appli research laboratori depart comput texa austin abstract report progress system driven applic network detect applic typic class problem coin monoton log monitor system system data sourc log transact assur servic support mileston success layer venu languag execut oracl venu abstract machin interfac data abstract oracl nativ trigger mechan identif system interest applic class enabl limit coupl mode identifi effect layer architectur introduct investig activ technolog detect network activ monitor hacker attack militari perform central command privileg comput subnet sniff network traffic initi set execut analyz traffic captur suspici packet discard obviou connect synopsi result filter subset connect log forward central command log oracl program sql period wake analyz activ event alarm condit human network analyst possibl notifi action network analyst exploit arrai queri tool investig track individu event watchdog effort encod central command compon directli rule execut rule tight integr exist platform worth note specif deploi system larg expect encod lead easili extend system pend deduct abil autom task conduct human expert watchdog effort integr rule system legaci consist primari task hook rule compon nativ sql standard trigger mechan select rule coupl mode implement refin ami layer applic challeng function provid layer determin perform bottleneck inhibit perform requir form basi identifi construct possibl impact sql standard research fund expert system project appli research laboratori texa austin sponsor air forc inform warfar center engin analysi contract watchdog compon expert system autom secur incid measur system watchdog typic larg class term monoton log monitor system system process form data log central primari reason dbm chosen storag method exploit servic watchdog receiv connect log suspici packet store monoton support highlight natur log exploit restrict number coupl mode simplifi transact interfac element monoton log monitor system class activ applic exampl rang point medic patient process monitor applic move support process increasingli time analysi success implement watchdog demonstr step technolog properti rule system familiar rule eca rule evalu typic modif recognit rule condit typic execut queri content condit action rule typic modifi extern side effect relat properti monoton log system necessarili monoton relat defin properti relat system borrow terminolog defin relat relat visibl rule system work relat rule system relat relat distinguish monoton relat relat defin monoton relat relat modif visibl rule system append consequ definit rule compon oper commit data rollback insert violat condit leav open possibl monoton relat purg modifi offlin mainten monoton relat includ connect log file network monitor applic custom order histori relat relat relat consist tupl modifi delet normal system oper exampl includ connect user relat inventori level relat distinguish monitor relat poll relat monitor relat relat rule compon automat notifi assum nativ trigger mechan poll relat stream updat rule system distinguish activ relat passiv relat activ relat eca rule mention relat event activ relat trigger condit evalu defin monoton log monitor applic activ relat monoton coupl mode relationship eca rule transact seri coupl mode defin relationship widom ceri summar class coupl mode coupl mode describ relationship occurr event evalu rule coupl mode describ relationship condit evalu action execut claim set coupl mode prune applic rule process necessarili begin insert activ monoton relat rule system oper commit coupl mode decoupl term origin rule assum singl origin rule quickli concern sql standard trigger commit semant oracl fire trigger uncommit counter basic undesir activ system result incorrect behavior includ origin rule execut due abort condit evalu data ad uncommit updat turn oracl facil maintain buffer watchdog successfulli avoid pitfal sql trigger semant address rule enabl event relat descend origin rule remain coupl mode determin ca coupl mode ec coupl mode rule event activ term chain rule discuss ca coupl mode correl mirank explor correct program execut environ categori isol guard guard stabil cursor greatest amount amount isol user mode dictat row access condit evalu action execut mode ensur transact affect rule system serializ mode properti guard stabil mode mode guard stabil stronger serializ impli lock gain condit evalu maintain action execut decoupl ca coupl mode venu system environ clip op oper individu rule atom impli action execut immedi condit evalu ca coupl mode dictat leav ec coupl mode chain rule recal event trigger rule evalu necessarili rule perform activ relat featur venu system constrain choic coupl mode semant venu treat rule atom action postpon action complet ec mode applic venu system support concurr execut strictli serial rule execut sequenc requir choic defer decoupl requir rule engin enforc desir rule order defer mode isol number durat lock decoupl mode potenti concurr reduc amount isol burden programm correct consist issu chose implement watchdog decoupl ec mode note multipl origin rule enabl insert event concurr execut origin rule rule concurr achiev execut multipl independ instanc rule engin rule system relat complet separ anoth buffer detail applic construct layer oracl underli due restrict oracl rule compon exist process address space unfortun situat singl piec inform simultan exist place place inform oracl disk storag area oracl memori address space trust oracl adequ manag buffer concern place rule system memori space perform rule system partial buffer relat impli familiar problem consist lock advantag properti individu relat maxim buffer possibl improv perform condit evalu relat exist address space rule system maintain main memori swap disk virtual memori subsystem monoton monitor properti lock addit occur notifi addit provid easili maintain buffer form activ relat venu languag leap match suit tupl time insert relat buffer access relat type impli lock queri venu activ extens famili expert system languag commonli implement rule modif extent yield rule evalu evalu condit execut action significantli costli extend venu provid eca rule expert system languag watchdog exploit function monitor monoton connect log insert event relat turn event monitor net result amount durat lock fewer access execut extend venu issu command abstract machin interfac instruct set leap program ami encapsul function common interfac heterogen compon applic architectur watchdog implement minim modif exist applic data flow figur illustr data flow watchdog data enter system bank monitor network insert oracl data oracl trigger sql trigger action trigger code special program escap mechan send data program program pass data socket watchdog compon data insert fifo queue data enter watchdog ami mirank instanc ami code remov data fifo place respect venu watchdog rule compon modifi add data oracl instanc ami eca rule subtract ca rule ca rule evalu modif eca rule evalu state event extens subtract modif includ time tag field append tabl definit oracl provid common commerci product provid facil sql standard regul program languag escap function cover standard trigger code unit elimin copi insert tupl data flow control flow bank network program fifo watchdog watchdog rule set evalu oracl db ami wrapper code trigger code network data event detect pass data program insert data fifo wake suspend thread ami call detect data fifo ami wrapper code evalu event db updat figur layer architectur control flow figur illustr control flow watchdog data enter system bank monitor network oracl updat oracl evalu activ monoton relat pass control applic trigger code trigger code spawn program send data socket trigger code transact side maintain decoupl coupl mode watchdog compon detach thread monitor socket data insert detect data fifo data remov share fifo queue insert activ instanc deem safe ami call modifi start process note address rule execut occur parallel tightli layer approach semant batch orient venu program encod coupl mode code modul code unit ad implement layer refin ami definit code commun oracl implement detach thread insert data fifo buffer oracl trigger code code pass data program code escap program pass data socket call interfac compon written program written sql trigger commun system socket fifo data structur manual implement futur compil extend output unit code recognit updat relat effect place read fifo buffer notifi venu activ relat updat complex time maintain ca coupl mode consist venu languag semant layer activ top legal implement implic implement quickli manifest implement lowest prioriti rule root modul rule read data fifo buffer updat system possibl exhaust rule lowest prioriti rule root model effect implement coupl transact start read buffer transact commit fire rule program correct ensur monitor relat append impli evalu improp data method constraint lead unwieldi larg buffer trigger event priorit bound priorit origin trigger insert data fifo buffer priorit origin read origin rule chain rule oper singl transact model maintain coupl mode program correct time transact process updat rate queu model establish execut schedul asynchron oper system hard constraint al correl mirank evalu commit standard option trigger semant notifi insert abort evalu rule incomplet relat util implement method investig altern softwar paper present implement layer system class applic coin monoton log monitor system layer advantag nativ trigger mechan ami implement watchdog demonstr clear expand sql trigger semant includ standard program languag escap order elimin addit copi object order elimin spuriou rule execut abort data rule evalu outdat relat substanti construct requir correct behavior applic examin numer rang coupl mode present literatur applic experi class subsum larg portion coupl mode present class preliminari perform result demonstr layer execut consist expert system data workshop speak result watchdog suggest consequenti simplif enabl practic system built todai bibliographi jc al approach modular intern confer tool artifici ieee dp mirank venu rule intern confer inform knowledg novemb transact load workshop activ dart novemb cl forgi user manual technic report depart comput carnegi mellon juli jc clip user guid artifici intellig johnson space june mccarthi architectur activ manag acm sigmod intern confer manag data pp dp bj perform lazi match product nation confer artifici juli system specif verif acm transact softwar engin vol octob pp dp overview activ intern symposium cooper system advanc applic decemb jeffrei principl system vol freeman decemb ceri activ trigger rule advanc process morgan kaufmann san 
ML/2379 6 ai magazin special issu uncertainti summer introduct special overview recent develop bayesian problem solv techniqu peter haddawi system artifici intellig lab dept intellig system lab faculti scienc technolog assumpt abstract year surg interest techniqu bayesian theori address problem ai theori norm framework repres reason problem uncertainti context research uncertainti ai commun develop comput techniqu build ration agent suit engin knowledg base special issu review recent research bayesian techniqu articl cover topic infer bayesian qualit theori provid introduct bayesian network cover applic bayesian model construct structur learn graphic probabl model past year interest tremend progress develop bayesian techniqu build problem solv system long uncertainti ai workshop found event precipit larg part fact mainstream ai commun time consid probabilist approach impract build intellig system workshop confer uncertainti attract contribut research broad arrai includ oper scienc concept bayesian comput techniqu develop uncertainti ai commun found mainstream ai routin paper primarili uncertainti area includ natur languag robot machin learn sign import regard bayesian techniqu build analyz softwar system fact journal acm recent introduc track comput bayesian concern ration behavior bayesian theori choic situat set act probabl distribut set state world outcom act state util function outcom space optim act maxim expect act physic speech delib complex plan kind action theori interest ai norm theori design agent capabl reason act condit uncertainti norm express form theorem state agent prefer obei set intuit appeal exist probabl function util function prefer action maxim expect util theori provid comput mechan build ration agent suit engin knowledg base issu primari focu contribut work conduct uncertainti research ai introduct discuss issu manipul probabl model provid discuss bayesian network cover applic bayesian model construct structur learn graphic probabl model research begun explor issu regard util model articl jon doyl thomason jim blyth follow provid discuss work interest introduct omit discuss excit valuabl research contribut point interest reader confer uncertainti ai workshop ai statist web page associ uncertainti ai read bayesian network bayesian network formal singl develop progress build practic system capabl handl uncertain inform book bayesian network publish ten year ago text book jensen al bayesian network direct acycl graph repres probabl distribut node repres random variabl arc repres probabilist correl variabl type path lack variabl probabilist independ quantit probabl inform form condit probabl tabl node tabl specifi probabl state node combin state parent tabl root node uncondit probabl kei featur bayesian network fact provid method probabl distribut set local distribut independ semant network topolog specifi combin local distribut obtain complet joint probabl distribut random variabl repres node network import consequ naiv joint probabl distribut tabl requir number valu exponenti number variabl system interact random variabl bayesian network drastic reduc eu number valu requir effici infer algorithm exist work transmit inform local distribut work full joint distribut separ qualit influenc variabl numer quantif strength influenc advantag knowledg engin build bayesian network focu qualit structur domain focu quantifi influenc complet specif joint probabl distribut common comput perform bayesian network determin posterior probabl random valu variabl network symmetr natur condit comput perform diagnosi predict common comput comput probabl conjunct set random comput combin valu random variabl comput piec evid influenc influenc hypothesi detail discuss bayesian infer techniqu articl bruce issu influenc diagram gener bayesian network analyz action addit chanc node share benefit bayesian network packag compani contact baron kc associ system dx solut seri knowledg expert softwar corp commerci bayesian network packag packag contact belief russel bn toolbox kevin patrick uc berkelei bug mrc unit imperi colleg school medicin ideal rockwel intern fabio paulo prakash kansa microsoft theori adapt system group universit de symbol probabilist infer bruce system univ pittsburgh yang regina free bayesian network packag practic bayesian network build problem solv system spawn small industri produc softwar build perform comput net tabl list commerci packag free demo version download packag packag free charg greatli reduc price academ user home page compani list tutori bayesian network archiv network tabl list freeli packag inform bayesian network packag web site applic bayesian model infer techniqu greatest bayesian problem solv techniqu wealth practic applic develop recent year sampl area intellig user inform autonom vehicl weapon medic diagnosi nice collect paper applic bayesian march special issu cacm singl wide distribut applic bayesian infer techniqu microsoft offic bayesian system offic suit applic offic assist base prototyp develop project lumier theori adapt system group microsoft research goal project develop integr comput system user model continu infer user goal user queri approach develop bayesian user model captur uncertain relationship goal user observ program sequenc action word user queri observ continu input bayesian model probabl distribut user infer system infer likelihood user receiv assist present moment ongo futur research project includ learn bayesian network model user log sourc event inform data autom vision dialogu obtain inform user goal applic develop theori adapt system group includ troubleshoot worldwid web vista mission control center johnson space center team flight control work monitor control space shuttl orbit subsystem team make control situat project vista project initi develop techniqu provid support flight manag complex inform displai vista develop research rockwel palo alto research lab stanford work close collabor expert system flight engin rockwel space oper compani system nasa mission control center houston year system bayesian network interpret live telemetri advic likelihood altern failur space shuttl system list problem order likelihood separ critic system model control level detail displai direct flight control import inform softwar develop johnson space center integr idea vista varieti monitor program instal mission control center martin martin marin system california artifici intellig center palo alto jointli develop autonom control logic system demonstr unman underwat vehicl develop unit state navi goal project develop softwar capabl control plan unanticip event manner minim risk vehicl loss maxim probabl success complet mission object acl system monitor progress analyz health detect analyz event impact mission make action modifi mission plan current longer achiev acl architectur hybrid bayesian compon compon perform unanticip event reason bayesian network model exist vehicl capabl uncertainti state capabl select altern unanticip event aim maxim achiev mission object bayesian ship tactic engin schedul ship system complex problem due low trajectori modern make detect close rang mechan constraint system interact bayesian tactic engin softwar weapon schedul design resid insid ship system tactic engin account uncertainti sensor measur threat identif error bayesian network determin optim time fire asset evid ship sensor limit constraint implement plan asset optim manner conflict occur weapon system fire target ideal plan transform planner resolv conflict produc optim work physic constraint asset microsoft pregnanc child care microsoft health product unit onlin consum health inform microsoft pregnanc child preview microsoft network model construct commonli occur symptom children run model select base chief complaint expert modul repeatedli determin question tailor multimedia present child health issu knowledg palo ca work theori adapt system group microsoft research develop test knowledg base infer procedur independ clinic test perform group collabor physician affili washington model develop microsoft bayesian network model infer bayesian pathfind bayesian network base expert system provid assist identif disord tissu section pathfind project stanford pioneer technic practic issu larg bayesian network success pathfind led commerci constel system set includ bayesian model node patholog addit bayesian model tissu repres kei area realm surgic patholog initi node model reason node probabl modul creat plausibl base featur enter system point diagnost user system identifi featur distinguish compet cost benefit observ test modul integr bayesian network patholog diagnosi librari slide evalu diagnost accuraci work assist node modul conclud work system produc significantli correct work system assist base inform integr capabl bayesian model node diagnosi hundr system world model construct structur success bayesian network li larg fact formal introduc structur probabilist model cleanli separ qualit structur model quantit aspect recent work attempt carri theme bayesian network technolog involv build larg network repres domain larg engin standpoint larg domain model collect independ smaller system address broad rang assembl model compon dynam lack modular make model difficult limit bayesian network essenti proposit node repres proposit variabl express gener relationship concept potenti instanc advanc system face broad rang dynam situat research endeavor address problem augment bayesian network concept program languag knowledg step direct repres class bayesian network set probabl repres node set parent condit probabl tabl free variabl permit express relationship class individu earli work line produc algorithm construct bayesian network knowledg base goldman goldman wellman algorithm capabl produc small tailor specif infer result comput save model evalu work provid formal semant knowledg base languag proof sound complet process construct bayesian network perform infer model construct approach appli problem militari situat assess student model intellig tutor synthesi data analysi program research structur introduc concept languag languag languag provid support structur model term interact build reason domain model level abstract learn graphic probabl model difficult time consum part task build bayesian network model probabl quantifi probabl deriv sourc obtain interview domain expert elicit subject probabl gather publish statist studi deriv analyt transmiss gene parent children learn directli raw data learn bayesian network graphic probabl model activ area research uncertainti ai commun recent year excel tutori learn bayesian network data friedman discuss larg addit learn learn structur bayesian network learn network structur point interest relat causal applic simpli learn human provid network structur classifi learn bayesian network model data complet incomplet structur network unknown discuss touch case structur straightforward case network structur complet data variabl network prior assum network paramet updat data bayesian learn dirichlet distribut commonli prior model paramet special case random variabl state binomi dirichlet express larg rang probabl function mathemat properti make calcul posterior distribut prior easi dirichlet distribut natur interpret term underli sampl size distribut obtain estim prior domain past experi estim base patient case assum model paramet independ dirichlet paramet updat independ data learn network incomplet reason valu simpli miss learn medic diagnost symptom patient distinguish valu miss random valu miss systemat valu systemat miss test run readili observ symptom present classic approach handl systemat miss valu build prior model data miss updat model observ data incomplet data lack observ variabl network hidden variabl make learn task easier data requir network variabl observ assum data miss techniqu popular gibb sampl handl continu domain variabl depend paramet gibb sampl stochast method approxim function initi joint distribut provid condit met distribut sampl state initi state satisfi joint distribut instanti chosen infinit condit met iter variabl averag sampl function approach expect respect probabl number sampl infin algorithm search maximum posteriori estim model paramet em algorithm iter expect step maxim step expect suffici statist miss entri comput bayesian network infer algorithm perform step expect suffici statist actual suffici statist mode paramet probabl network structur paramet maxim em algorithm fast disadvantag provid distribut model paramet stuck local substanti amount data miss unknown structur common approach learn structur paramet data introduc score function evalu network respect train search network metric obviou choic bayesian posterior probabl network observ data score difficult comput altern criteria typic commonli metric belief score function heckerman minim descript length base score function mdl score function prefer network fit data simpl approxim bayesian number case mdl score bayesian assum dirichlet distribut uniform prior structur mdl belief score function likelihood function measur network fit observ data data independ network structur likelihood function product term modular evalu candid network local evalu chang remain chang part network data longer likelihood perform infer evalu em algorithm gradient descent step em algorithm requir comput probabl event instanc train ineffici make matter local chang part network affect evalu chang part neighbor network visit evalu requir call em procedur make singl chang current candid network friedman introduc innov perform search structur insid em procedur current estim unknown distribut complet data procedur work effici complet data approach maintain current network candid iter attempt find network structur comput expect statist evalu altern search complet data exploit decomposit properti score metric algorithm appli score function approxim bayesian mdl recent friedman extend approach work exact bayesian score evid exact bayesian score assess gener properti model data principl incorpor prior knowledg learn process articl special issu introduct provid mention rich arrai techniqu infer bayesian network articl bruce detail discuss bayesian briefli describ aspect varieti exact approxim infer techniqu mathemat foundat theori concern ration choic plan natur applic bayesian techniqu articl jim blyth show plan extend classic ai plan outlin central issu describ altern approach build planner articl jon doyl thomason survei previou accomplish field discuss futur direct develop problem solv system argu quantit techniqu tradit theori proven fulli adequ support attempt artifici intellig autom make qualit approach call provid overview fundament concept discuss qualit pointer recent work direct acknowledg vu eric mark shavlik comment draft paper work partial support nsf grant refer bayesian air forc institut technolog artifici intellig ohio bayesian network heckerman whittak seventh intern workshop artifici intellig statist morgan kaufmann san electron version kanazawa adapt probabilist network hidden variabl machin learn js construct belief network comput van transform system nasa proc intl workshop softwar transform lo guid literatur learn network data ieee transact knowledg data buntin oper learn graphic model journal artifici intellig buntin theori refin bayesian network proc commun special issu applic bayesian wellman march jm expert system probabilist network model springer cooper bayesian method induct probabilist network data machin learn rubin maximum likelihood incomplet data em algorithm roi statist soc mj lc van der fv jensen work note workshop build probabilist number friedman bayesian structur em algorithm proc fourteenth confer uncertainti artifici friedman learn belief network presenc miss valu hidden variabl proc fourteenth intl conf machin friedman goldszmidt slide tutori learn bayesian network data vanlehn procedur gener hint bayesian network student model proc rp goldman charniak languag construct belief network ieee transact pattern analysi machin rp goldman charniak dynam construct belief network proc sixth confer uncertainti artifici haddawi gener bayesian network probabl logic knowledg base proc tenth confer uncertainti artifici heckerman tutori learn bayesian network technic report microsoft washington march nov dm chicker learn bayesian combin knowledg statist data machin learn norm expert part pathfind project method inform mc pool dynam approach probabilist infer bayesian network proc sixth confer uncertainti artifici horvitz person commun jan bayesian user model infer goal softwar user proc fourteenth conf uncertainti barri displai inform make proc eleventh confer uncertainti artifici ra howard je influenc diagram read principl applic ra howard je vol strateg menlo fv jensen introduct bayesian network springer koller structur probabilist bayesian network proc koller pfeffer bayesian network proc thirteenth conf uncertainti pj learn probabilist network lam bacchu learn bayesian belief approach base mdl principl comput intellig sl lauritzen em algorithm graphic associ model miss data comput statist data martin autonom control logic guid unman underwat vehicl press martin missil space commun palo april bayesian reason autom assist kb construct situat specif belief network proc fourteenth conf uncertainti lehner schedul uncertainti ship submit ieee special issu intellig bayesian ship tactic engin bn mc sp evalu expert system node patholog human patholog probabilist reason expert system john wilei haddawi answer queri probabilist knowledg base theoret comput haddawi probabilist logic program bayesian network knowledg lectur note comput vol springer pearl probabilist reason intellig system morgan san print pool probabilist horn abduct bayesian network artifici pool repres bayesian network probabilist horn abduct proc seventh confer uncertainti artifici project manag displai complex inform db rubin estim causal effect treatment random studi kanazawa local learn probabilist network hidden variabl proc fourteenth intl joint conf savag foundat statist john wilei edit publish dover dj spiegelhalt sl lauritzen sequenti updat condit probabl direct graphic structur proc fourteenth conf uncertainti artifici gf cooper moral morgan san js rp goldman knowledg base model knowledg engin 
HCI/1893 4 introduct adapt web inform disabl elderli user alfr kobsa gmd fit german nation research center inform technolog schloss st germani substanti research standard effort exist make easier peopl physic impair perceiv interact web page paper describ work aim cater content web page includ elderli peopl user vision motor impair system relat effort project discuss experi report world wide web frequent visit electron resourc access ramp electron inform highwai millennium web access ideal order creat econom dispar societi special effort put make access web includ peopl disabl elderli peopl recent minor user inform technolog term coin denot design methodolog comput interfac automat preclud user special case softwar develop date effort made render web access user sensor defici primarili includ user visual impair morlei al extent user motor impair al adapt research includ altern input output modal speech adjust behavior user interfac element toolbar rotat navig web page make interfac interact interfac access render web site fulli access user respect percept respect inform convei web user disabl specif inform share user inform access interest inform import user inform music event deaf amount addit inform requir user group burden user inform present paper discuss principl appli principl inform present addit appli interfac interact interfac approach consist adapt inform individu inform express user model user research project al develop prototyp tourist inform system wide rang includ peopl disabl elderli user special inform area wide varieti inform observ stem prefer physic abil individu user user analysi carri context reveal special inform peopl requir substanti inform user disabl group inform redund user omit present inform detail inform access build exist dimens ramp type width type floor import automat provid warn rout present insurmount obstacl step slope gradient rout difficult master rout gradient rout plenti blind modal present inform chang tactil audio supplementari orient navig addit tabl content intern extern link hypermedia al automat warn presenc mobil acoust long distanc conquer tactil acoust main mobil barrier user group elderli user includ categori handicap due ag health condit receiv respect special inform automat warn presenc narrow spiral surfac hand long distanc conquer rest relief point elev mark special inform stem disabl inform distinct user user interest specif subject area automat receiv detail inform user lack interest comput usag instruct augment explan experienc comput user user network access slow inform high data volum video replac order reduc download time time hypermedia system extrem critic usabl point view order cater inform techniqu area adapt hypertext hypermedia system al proven system dynam gener hypermedia page base constantli updat assumpt user store user model overview system tourist inform system adapt web page individu present user system hypermedia inform metropolitan area public varieti user travel agenc elderli blind user slight form abil system access travel public inform kiosk hardwar softwar network surround vari wide order cater user usag exploit method tool develop context adapt adapt system user model adapt hypertext hypermedia system al consist main multimedia inform domain modal includ restrict data model user model server host model user model primarili inform abil extent assumpt interest rang circa result muscl impair muscl inform resourc control structur repositori gener web page includ link entri rule base current user model irc page encod extens html adapt gener hypermedia document descript basi irc current user model user interfac present hypermedia page perform adapt level user select modal interact form acquir assumpt user base sourc short initi interview acquisit primari assumpt al user valuabl sourc inform initi assign user user subgroup basic inform disabl store read distanc cm depend electromagnet surround inform visual manual adapt modal interact interfac user approach system fig show fact current user mildli store read antenna monitor inform automat lead screen magnif base primari assumpt user addit inform applic system draw infer order acquir assumpt user user request detail inform histori assum interest detail inform henceforth automat provid stereotyp assumpt interest characterist user subgroup presum knowledg domain expert presum interest precondit stereotyp activ specif mean assumpt stereotyp assign user figur automat screen magnif mildli user read user profil distanc web page depict fig inform facil amen hotel itali page separ area left column background main menu central area inform servic offer column includ menu button point relev inform present page tailor user experienc comput system special inform peopl initi questionnair option navig path displai upper left corner experienc web user usag requir mental model hypermedia space hand path left web novic navig tool restrict hierarch siena menu side user altern navig element hypermedia page hand icon point link lower corner point inform interest current user assumpt acquir system infer mechan base navig histori user request detail inform cook style assum interest restaur figur adapt adapt element web page option element button toolbar browser upper corner button avail button user button press fig user automat obtain addit inform select user group addit inform peopl part page descript includ browser html parser process page behav web access commerci browser extra button remain html page culti walk button design peopl inform behalf disabl person travel agenc clerk custom work system technic point button activ deactiv stereotyp tabl present user attribut system deem relev current user model user modifi attribut select request adapt version tabl includ attribut fig show adapt version user attribut subsequ confirm experi system subject extens summ evalu citi scenario total subject al user busi travel agenc user vision motor impair common question share experi develop system benefici specif technic feasibl end user introduc adapt adapt inform content user interfac hypermedia inform system user subject usabl test determin effici satisfact specif contribut adapt adapt data collect questionnair analysi log file result conclud adapt featur gener appreci benefit inform system li user satisfact effici gain error reduct gener find experienc user inclin advantag adapt featur user experi adapt featur recogn understood frequent expert adapt comput web expert path navig travel agenc clerk possibl creat shortcut travel agent appreci button button gener correctli frequent user gener satisfi inform provid system attitud user found system properli design inform access clear mobil appreci level detail reach system enthusiast user easili find inform inform present concept mobil differ impair user properli reflect system refer al global evalu experi act brussel home page user requir brussel al adapt hypertext hypermedia system kluwer academ publish al adapt adapt inform provis includ disabl elderli peopl review hypermedia multimedia prototyp www browser special adapt xv ifip world comput comput assist technolog pp al prototyp browser special adapt ercim workshop interfac access stockholm al web access blind visual impair comput user technolog educ visual handicap acm june user recent prospect hazard adapt user principl practic amsterdam al auditori navig design evalu hypermedia system blind user asset annual acm confer assist technolog marina del art navig hypertext commun vol adapt user support ergonom design manual automat adapt softwar lawrenc erlbaum publish al user model acquisit heurist base dialogu act intern workshop design cooper user model stereotyp cognit wai homepag acknowledg research partli fund european commiss act programm carri collabor european partner 
ML/480 6 converg theori applic factor distribut algorithm heinz muhlenbein theoret foundat gmd laboratori gmd forschungszentrum st augustin abstract paper investig optim addit function evolutionari algorithm call factor distribut algorithm fda base factor distribut gener search point separ adf consid map gener linear function metavari defin multipl allel map transform fda univari margin frequenc algorithm exact equat select comput assumpt select truncat select approxim equat time converg deriv analysi function fda numer investig separ function time converg similar separ adf fda outperform genet algorithm recombin string keyword fisher addit genet factor distribut introduct genet algorithm popul base search method set point point select point gener genet oper recombin mutat simpl genet algorithm select point denot vector discret random variabl distribut gener averag fit popul simplic assum binari variabl select scheme call averag genotyp proport fit valu select guid creation gener point discret densiti defin paramet mean straightforward implement equat prohibit central question genet algorithm popul base algorithm formul equat approxim substanti exponenti comput real world comput partnership extend genet approach base probabl theori approxim equat explicit aggreg structur aggreg schema give schema extend usual notat defin schema valu held variabl free sum equat member schema obtain probabl select probabl term margin distribut abbrevi schemata defin margin distribut state question term probabl set margin distribut exist good approxim equat comput polynomi simplest choic order schemata univari margin distribut univari margin distribut algorithm point gener densiti order schema defin gene loci distribut equat order schemata give bad approxim claim schema theorem creat point recombin good approxim equat muhlenbein numer theoret evid recombin behav similarli approxim equat equat order approxim complex distribut higher order schemata schemata illustr problem loci order schemata gener exist close express function bivari distribut problem discuss detail muhlenbein al paper fit function factor product margin distribut addit function ae ng ng outlin paper investig separ section transform function function metavari defin multipl allel problem exact equat select comput equat fisher fundament theorem natur select gener linear function section weak form fisher theorem proven arbitrari fit function examin relat structur equat structur fit function result comput approxim solut linear fit function section investig special function defin set overlap variabl condit distribut section deal truncat select approxim solut obtain defin evalu factor distribut algorithm numer exampl fda extens factor map metavari multipl allel assum probabl express product lg index set disjoint ng combin metavari allel js fit function formul fit function metavari valu map js demonstr map exampl ff ff ff ff defin metavari function transform ff ff ff ff ff ff ff ff ff metavari obtain gener linear function higher order interact consid ij ij ij ij metavari larger combin bit metavari assum index set ij disjoint simplic assum set suitabl map transform gener linear function map function mg set mod div mod div xm defin metavari defin equat fit function transform gener linear function lead larg set reason partit set separ metavari partit case transform function linear comput section equat select gener case exact equat call mind domain definit discret vector discret random variabl shorthand notat element proof variabl denot variabl function denot margin probabl distribut polynomi order distinguish interpret write note free addit constraint deriv respect addit definit definit subset denot product space jj sum fix ji multipli formal proof binari gene comput exact equat select express defin addit genet varianc summat equat equat obtain differ equat univari margin frequenc defin definit univari margin distribut algorithm gener point origin space gener point deriv map defin factor distribut algorithm origin space size index set js bound fix constant margin distribut comput enorm reduct compar factor theorem multipl allel select start taylor consid fix variabl notat defin structur proven part express deriv order higher involv calcul deriv shorthand notat paramet omit interpret variabl term independ term obtain theorem select arbitrari fit function separ fit function transform gener linear fit function function fisher theorem valid corollari gener linear function select combin equat obtain deriv independ higher deriv show complet linear fit function higher order deriv vanish gener corollari fit function order interact order deriv vanish yield order deriv vanish gener weak version fisher theorem fisher theorem valid gener linear function arbitrari function show greater equal point gener equat proof base gener form inequ baum written notat theorem ij polynomi nonneg coeffici homogen variabl ij fp ij point domain ij ij denot point coordin ij ij ij ik ik theorem select margin frequenc remain function variabl consid variabl defin ij constraint ij fulfil show equat ident equat univari margin frequenc easili obtain ij defin equat ik valid margin frequenc fulfil assumpt theorem numer properti averag fit popul properti termin algorithm section investig exist vanish higher order deriv structur relat paramet consid fix omit formula theorem jj subset exist function proof induct jj lengthi omit theorem state number interact variabl order higher deriv vanish give exampl assum deriv gener linear function binari variabl conjectur assum deriv form ij section approxim solv differ equat linear function linear fit function theori obtain equat chang allel frequenc linear fit function arbitrari alphabet arbitrari linear function ff ff ff ff margin frequenc step select equat ff ff theorem linear function margin distribut select select ff ff ff sort equat multinomi fix substitut order obtain multinomi chang sign product involv lead ff ff ff evid ff ae jj term ff occur jj time altern jj jj obtain ff yield ff ff ff deriv order higher final ff ff ff ff ff differ equat easili comput numer analyt solut complic deriv approxim solut special case theorem ff ff approxim margin frequenc ff ff ff ff ff ff ff ff ff ff fi ff obtain equat ff approx figur approxim exact allel frequenc calcul frequenc ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff fi ff ff ff ff ff ff ff fi ff ff ff ff ff fi ansatz obtain equat ff ff ff ff ff ff ff fi divid equat function defin ff case obtain exact solut equat approxim exact difficult function ff ff fit function theorem comput numer valu numer compar approxim valu exact valu result shown figur plot frequenc logarithm lead linear graph predict condit probabl distribut index set analysi difficult probabl distribut longer product independ distribut introduc condit probabl order describ problem simpl index set probabl distribut factor jx factor product probabl constraint determin paramet gener distribut variabl paramet slight reduct number free paramet goal calcul omit paramet clear context equat obtain equat suitabl aggreg deriv differ equat distribut defin probabl formul result theorem theorem select fit function set give factor jx obtain jx jx chosen consist actual sum equat jx lead procedur insert yield jx jx denot univari margin frequenc set jx jx jx jx jx jx lead differ equat note result formal similar term jx complex usual term tediou calcul final express deriv conjectur formal obtain exact equat simpl overlap set analysi extend chain overlap variabl interpret calcul condit term difficult set overlap irregular difficult obtain factor probabl investig problem numer simul approxim deriv theori factor distribut algorithm fda base factor distribut gener form assum jy ae ng mathemat discuss factor problem reader obtain factor exampl discuss detail set transform fit function gener linear function select suit mathemat suit practic optim algorithm select weak final stage algorithm fda truncat select fda ffl step set gener ae point randomli point equat ffl step select ffl step comput condit probabl jy select point ffl step gener popul jy ffl step termin criteria finish ffl step add point previou gener gener point ffl step set step fda extens work origin space separ function algorithm gener search point fda combin mutat recombin genet algorithm oper probabl distribut factor probabl defin equat faster proport individu gener local approxim condit margin distribut jy zn arbitrari larger yield uniform distribut chosen small jy fda micro macro structur subset disjoint fit function gener linear function macro variabl defin integ select analyt deriv exact differ equat margin distribut truncat select comput exact differ equat show similar fda section simul converg result truncat select analyz simpl fit function detail count number bit string approxim relev fda recal theoret result al assum loci ident univari margin distribut initi popul differ equat obtain equat select denot varianc popul denot select intens obtain differ equat differ equat differenti equat deriv solut sin equat complet describ dynam complet show select truncat threshold select intens tabl tabl select intens comput gen number gener converg converg achiev averag fit popul equal fit gen obtain set gen gen depend size size popul truncat threshold result expect fda behav truncat select ffl gen ffl gen conjectur statist properti confirm conjectur valid popul size larger critic popul size defin minim popul size find optimum high determin critic popul size difficult succe analyt formula numer result test conjectur number gener equilibrium function function set valu function order ident set valu theori expect result gen equal gen smaller fda main altern rest gen ga tabl gener truncat threshold result tabl confirm predict note equat predict gen obtain actual simul fda ga genet algorithm truncat select uniform crossov slightli gener mention function genet algorithm gener knowledg turn fit function test suit denot number bit turn dec defin dec dec defin separ decept function order dec factor distribut function separ function function function defin function function chain interact structur variabl set function difficult optim global optimum optimum trigger optimum function function factor numer comparison separ function system ij ij sum spatial neighbor ij factor grid spin sequenti number till js js distribut variabl variabl left side condit sign grid interact cover factor class model exact solut comput comput fairli difficult free boundari problem tabl numer result order shorten comput run stop occurr individu maximum number gener gen gener count stop result gen depend great differ fit function larg differ separ number variabl consid separ function initi margin probabl speed converg function equat obtain good predict gen gen fda fda ga fda fda fda gat fda fda fda gat fda fda gat tabl numer truncat threshold gen remain constant popul size larger critic popul size shown entri tabl model complex optim problem reflect popul difficult greater popul size model small popul critic popul size larg function critic popul size dramat expect comparison note result genet algorithm ga uniform crossov truncat select genet algorithm solv separ function difficult function solut model surprisingli simpl ga found optimum run case fda outperform genet algorithm qualiti solut obtain number function evalu obtain optimum factor distribut algorithm fda extens separ adf separ function behav function multipl allel select deriv differ equat margin frequenc truncat select confirm numer number gener converg estim equat deriv simpl linear function percentag initi popul gener gen smaller shown fda effici separ adf factor condit probabl differ converg speed optim separ separ function difficulti optim problem reflect popul size obtain optimum give impress fda prefer genet algorithm simplest fda implement set function solv surprisingli larg fda function complex gene interact structur success fda critic depend correct probabl model problem constraint optim problem discuss muhlenbein al refer le ja inequ applic statist estim probabilist function markov process model ecolog bull math genet theori natur select de genet algorithm optim machin learn adapt natur artifici system press scienc breed applic genet algorithm evolutionari equat select predict evolutionari distribut graphic model evolutionari optim submit 
Agents/1690 1 limit logic belief analysi antonio moreno superior universitat de spain abstract process ration inquiri defin evolut belief ration agent consequ intern infer procedur interact environ belief model formal doxast logic world model kripk semant provid intuit semant commit model agent logic omnisci perfect reason problem avoid syntact view defin arbitrari set sentenc proposit belief logic articl syntact view world dynam analysi agent belief suggest order model process ration inquiri agent perman engag compon logic dimens analysi perform modifi version analyt tableaux model evolut belief due agent infer power shown reason achiev agent deduct abil control restrict tautolog logic agent oblig perform exhaust analysi initi tableau document extend version limit logic belief februari introduct aim work model process ration transform belief intellig agent time classic philosoph tradit consid compon ration consist applic infer procedur actual belief addit infer belief discoveri incompat empir add remov belief result observ perform agent environ dimens shown section logic belief tool formal reason belief perform intellig agent logic modal logic necess oper interpret belief formula languag semant interpret world model kripk semant semant postul set world binari access relat connect world world real agent belief agent proposit world iff true world world st drawback appar logic model agent process ration inquiri main problem model agent logic omnisci classic perfect reason belief close classic deduct properti agent design model ideal agent children muddi children inadequ realist agent consid argument logic omnisci perfect reason ffl agent unawar fact halpern fact account logic gener ffl agent limit resourc time requir space perform evid ration agent implement real comput ffl agent ignor relev rule agent told rule modu view consid konolig phd thesi agent model base set belief set infer rule fund european human capit mobil project virtual contract limit logic belief februari goal overcom develop model process ration inquiri evolut belief ration agent time consequ interact world intern inferenti world model kripk semant natur intuit semant modal avoid logic omnisci perfect reason syntact view world defin world set formula proposit doxast languag condit partial fact actual world inconsist inconsist appar hidden deduct argu favour kind world number ffl logic inconsist world model instanc situat agent unabl belief account subset draw consist context inconsist belief consid remark shoham fagin halpern model situat logic local consid idea ffl defin interest procedur inquiri inconsist set belief shown ffl situat tautolog inform believ model instanc initi state inform rasa theori lock rousseau world situat conceiv agent world impli world fail tautolog logic omnisci deduct close agent perfect ai literatur author deal problem introduc imposs world logic connect behav usual tautolog inconsist formula hold imposs world world world situat semant approach allevi problem logic omnisci perfect problem detail review main inconveni syntact approach defin world set agent limit logic belief februari minim set infer capabl author argu make analysi agent belief rest document show belief evolv time consequ analysi dynam belief analysi evolut agent belief time due dynam analysi ffl logic dimens model inferenti process carri ffl dimens model belief interact agent environ rest document describ logic analysi descript side analysi pictur belief system found basic idea side analysi agent order corrobor falsifi result obtain logic fashion belief obtain logic analysi falsifi belief procedur chang initi belief agent proce logic analysi set belief logic analysi languag finit fix set basic proposit logic oper modal belief oper formula languag restrict standard proposit prefix negat possibl sequenc modal oper world defin set formula logic analysi agent gener sequenc access relat standard kripk defin set belief logic analysi model evolut agent belief time due intern deduct infer evolut illustr small agent belief world conjunct materi implic defin negat disjunct usual agent subset proposit modal logic consid restrict overcom futur work limit logic belief februari world defin set formula initi set belief agent gener initi access appli standard kripk semant world world iff formula believ agent belief agent world restrict proposit doxast proposit common access world belief ontolog world defin repres figur rs ff figur world doxast altern world repres rectangl left repres ffl formula ffl formula ffl formula believ ffl formula believ class world repres world class world class ff world logic analysi agent belief perform modifi version analyt tableaux method tableau divid left column analysi start tableau left column initi nest belief consid similar analyt tableaux tree level nest exampl initi set belief negat belief treatment tricki column includ rectangl reinforc interpret belief figur fourth column column ff differ respect classic analyt set formula limit logic belief februari set belief agent light previou argu repres agent belief descript world initi connect belief true world repres ff tableau repres class world proposit tableau true tableau repres world formula left column formula column explicit neg inform analyt tableaux enrich kind situat model standard world model semant formula classic pictur neg inform includ situat explicit inform consid model situat ffl agent thought ff ffl situat intermedi step complet stage ffl situat agent thought believ negat split rule analyt tableaux method modifi appli disjunct left column tableaux gener oe left oe left column left column oe column rule agent explor possibl account truth member true formula applic rule disjunct situat shown figur reach partit class ff ffl ff world ff ffl ff world ff note impli recal classic analyt tableaux analysi disjunct gener ff fi minor modif respect classic conveni talk formula open tableaux longer cumbersom formula branch tableaux tree limit logic belief februari sp sp sp figur split rule appli ffl ff world ff ffl ff world ff tableau repres class world ff repres ff repres ff world class ff consid analyt tableaux model initi set formula model set fail true tableaux analysi tableau close circumst ffl tableau formula negat left world belong class world repres tableau explicit consid logic imposs elimin analysi ffl tableau formula left column repres world repres empti set world dismiss analysi reason tableaux posit neg easier pictur analysi disjunct partit class doxast classic class world repres gener split rule world belong class world close condit classic method situat instanc replac limit logic belief februari left agent class world repres tableaux ff ff logic imposs rid world close dismiss branch logic analysi remain tableaux tree effect analysi formula restrict doxast altern shrink ff ff applic split rule model gener access restrict set world restrict shown figur rs sp sp ff ff ff ff sp sp sp ff figur reduct belief altern world class ff world class ff chang access relat impli chang belief agent belief point notic limit logic belief februari restrict belief altern addit formula belief includ world ff world ff includ world ff analysi show belief agent fact reflect figur addit fourth column agent pursu analysi belief gener access disjunct obtain result shown figur sp sp sp sp figur analysi intuit interpret restrict result partit ff subclass ffl ff world ff posit belief formula column class doxast neg belief formula column class doxast altern limit logic belief februari ffl ff world ff ffl ff world ff ffl ff world ff world ff world subclass ff ff ff ff ff ff ff repres tableaux situat depict figur ff sp rs sp sp ff sp ff ff sp ff ff ff ff ff sp sp sp figur reduct belief altern notic set posit belief formula common world common world set neg belief limit logic belief februari doxast altern world ff situat world class ff believ fact reflect fourth column rectangl repres figur disjunct open end analysi agent perform introduc instanc topic consid section introduc tautolog shown section classic analyt tableaux method modifi number ffl tableau set formula left repres set world formula left column formula column ffl split rule gener member disjunct left left column column repres world disjunct hold ffl appli split proposit formula tableau maintain result delet agent add tag proposit recal import modif classic analyt tableaux method agent add open tableau tautolog classic analyt tableaux tautolog introduc open tableau notic agent belief stage analysi formula open tableaux agent freeli introduc tautolog open logic omnisci avoid case agent reach deduct consequ initi set perfect reason split rule disjunct left rule order formula gener negat formula form column rest document initi posit disjunct split except explain formula ff logic consequ initi set belief obtain logic analysi belief introduc formula open tableaux exhaust tableaux close situat eventu reach set inconsist limit logic belief februari forbid tautolog drastic avoid logic omnisci perfect put sever limit agent deduct capabl definit formula deriv initi set iff perform logic analyt tableaux method section initi tableau left left column open tableaux recal open tableaux repres class doxast point logic analysi open tableaux formula formula doxast altern kripk belief agent point set potenti belief equal set deriv formula set definit set ffl fc st subsum ffl fc st ffl ffl lemma iff ffl formula ffl negat ffl inconsist condit trivial call tc rest formula automat set deriv belief creat tableau immedi close condit state remov gener inconsist set assum formula sake assum rest document formula initi set formula transform conjunct normal form open formula deriv inconsist consequ avoid model limit logic belief februari deriv iff tableau fc happen iff set inconsist tautolog set deriv formula small subset logic consequ initi set tautolog logic omnisci perfect reason unavoid fact suggest interest control agent deduct tautolog logic analysi instanc agent instanc axiom exclud tautolog natur agent believ formula negat except permit introduct formula split column agent explor altern logic analysi guid search exampl give side doubt possibl ad kind tautolog analyt tableaux idea tradit classic proof suggest hintikka interrog model inquiri model process scientif inquiri plai interrog game plai inquir natur inquir seek true perform deduct move analyt interrog question put natur interrog hintikka introduct instanc left side formula formula substitut instanc subformula formula respect introduct instanc control set formula believ agent set deriv formula depend kind instanc put restrict principl constrain set obtain formula set potenti assum restrict deriv defin definit agent introduc open tableau logic analysi instanc initi set belief logic logic analysi agent rule tableau close classic analyt tableaux simpli proof deriv immedi close left column hintikka classic analyt limit logic belief februari formula initi set iff perform logic analysi left column open tableaux set formula larger set deriv shown lemma iff tc ffl ffl agent trivial condit tc previou iff formula introduc open tableaux eventu tableaux formula left column formula set kind idea push tautolog consid preliminari studi instanc tautolog logic main point logic analysi logic omnisci perfect reason potenti belief agent control restrict tautolog agent logic analysi logic analysi suggest document introduc model imperfect reason classic analyt tableaux method order prove ff consequ set fail build model exhaust analyz tableau close analyt tableaux tree analysi import agent oblig exhaust analyz tableau model imperfect reason due lack previou kind imperfect model agent perfect deduct power agent decid stop logic analysi point deriv proce task make test order corrobor falsifi current belief obtain logic deriv agent extra tableau formula left column gener formula tableaux obtain analysi hintikka model set wait model agent limit logic belief februari summari futur work articl dynam analysi ration agent set belief logic dimens logic analysi perform modifi version analyt tableaux agent deduc fact initi set belief logic omnisci perfect reason avoid ffl deduct power agent control restrict instanc tautolog introduc open tableaux ffl agent decid perform exhaust logic intermedi belief good guid behaviour serv basi analysi futur work constraint studi discov tautolog yield interest set deduct consequ kei issu research studi interact strand analysi preliminari studi refer mathemat north semant entail formal van de van logic logic journal philosoph logic framework logic explicit comput intellig awar limit reason mit mose guid complet complex modal logic knowledg artifici intellig limit logic belief februari knowledg cornel world journal philosoph logic logic interrog model scientif reason knowledg paradigm epistem interrog model inquiri gener theori argument commun cognit introduct modal una la logica ed calculi construct institut languag dissert seri deduct model morgan semant analysi modal logic normal modal proposit fur mathematisch logik und grundlagen mathematik logic implicit explicit knowledg comput intellig dynam belief fourth intern colloquium cognit extend version research report depart technic belief research report human commun research de la vi confer spanish associ belief submit limit logic belief februari logic belief submit logic scientif hutchinson logic reason knowledg knowledg engin review logic varieti artifici intellig mathemat theori academ springer 
ML/573 6 rewrit logic metalog framework david basin manuel mesegu institut germani depart spain comput scienc sri usa abstract metalog framework logic methodolog repres logic reason metalog properti logic framework good metalog framework logic support ectiv reason theori initi model present concret realiz idea rewrit logic theori rewrit logic initi model logic support ectiv reason impli induct reason valid prove properti initi model theori rewrit ection reason metalevel properti uniformli ect induct principl prove rewrit logic theori extens show ectiv methodolog ectiv framework di kind formal prove relat theori establish properti class theori report implement induct theorem prover design base result present paper introduct logic framework logic methodolog emploi repres formal system number logic framework compar analyz distinguish intend applic distinguish logic emphasi reason simul deriv framework metalog emphasi reason logic relationship logic metalog framework includ abil reason logic entail adequ simul entail induct plai central distinguish logic framework metalog counterpart logic proof rule construct deriv entail approach logic framework isabel edinburgh lf formal logic theori induct present theori induct david manuel mesegu present theori framework support induct term proof theori metalog essenti induct theori standard argument requir induct formula deriv object theori induct essenti comput scienc reason oper semant ectiv metalog framework approach metalog framework motiv observ logic syntax proof view carrier set induct built syntax proof constructor logic framework metalog framework share common basi logic framework applic constructor suce simul deriv object metalog preserv induct natur algebra formal metalog initi model syntax proof formal object logic logic ectiv theori initi step logic framework metalog framework ect metalevel induct principl formal logic sum framework ection initi metalog make idea give concret realiz rewrit logic present standard deduct theorem minim logic rewrit logic candid ectiv metalog good rewrit logic demonstr good logic framework balanc point express natur formal di erent entail weak theori initi model mean sound induct principl reason respect model prove theori rewrit logic kei ect reason principl metalevel contribut theoret practic work contribut answer question metalog ectiv logic theori initi answer illumin interrelationship logic metalog ection kei ingredi turn logic framework initi model metalog provid evid rewrit logic metalog framework ing combin ectiv metalog framework nontrivi kind reason relat work approach consid past strengthen logic framework function metalog framework di er signi cantli logic basi ection metareason approach formal theori framework logic support notion modul explicitli equip induct principl theori formal collect modul type theori modul includ induct principl reason term proof approach power relat di erent theori formal altern approach formal theori directli induct de nition framework logic framework theori strong formal induct principl simpl theori fs carri experi formal fs induct de nition term framework induct rule reason term common choic formal theori induct de nition strong framework logic logic type theori calculu construct induct de nition logic set theori intern develop theori induct de induct de nition correspond term formal monoton de induct principl formal deriv framework logic calculu induct de induct principl simpli metalog current research area induct principl logic support abstract syntax organ remaind paper organ section present idea ectiv metalog framework abstractli formal requir metalog section present background materi rewrit membership equat languag section discuss induct principl membership equat theori present simpl notion membership equat theori section discuss rewrit logic logic david manuel mesegu section show combin initi ection rewrit logic metalog framework show uniform induct principl theori extens present section formal metareason rewrit logic metalog proof deduct draw section ectiv metalog framework section begin de ning ectiv logic base describ properti sucient ectiv logic framework function ectiv metalog framework ectiv logic ectiv logic logic import aspect theori repres reason logic gener axiomat notion ectiv logic recent notion express term gener axiomat notion entail system captur entail relat logic entail system consist class sign signatur sign speci es syntax function sen assign signatur sign set signatur sign entail relat sati es properti transit omit subscript clear context theori pair consist signatur sign set sentenc extend entail relat theori obviou de ning de nition entail system nonempti set theori theori call rewrit logic metalog framework entail system call ectiv ectiv logic logic entail system ectiv class nite present theori logic requir ectiv metalog framework requir logic framework function metalog framework section approach formal metareason past approach base ectiv reason abstractli requir logic framework weak valid induct principl reason logic framework express viabl logic logic framework ectiv note speci es requir framework logic altern formul abstract framework logic theori initi form induct reason valid prove sentenc respect initi model theori method equat induct initi concept explain requir list sucient turn logic framework metalog framework requir sati logic entail relat repres theori logic requir sati preserv induct natur algebra syntax proof logic repres argument requir induct formula deriv logic appli framework logic prove theorem logic theorem formal notion initi induct reason principl case di erent form equat logic oper ariti smaller regular cardin induct principl tran nite interest logic framework suitabl repres logic practic interest framework logic theori initi model induct principl david manuel mesegu sentenc initi model theori repres object logic prove induct deal requir relat di erent logic famili logic deduct theorem minim logic deduct relat di erent deduct prove obtain rst ad axiom prove mean sentenc formal relat initi model di erent theori ection plai logic framework sati es requir theori metalevel logic theori theori logic framework theori initi model kei exploit order formal relationship initi model object theori theorem initi model induct principl reason initi model object theori induct principl reason initi model theori give concret instanc idea case rewrit logic show class rewrit logic theori induct principl theori simpl induct principl reason induct properti theori reason induct theori induct reason properti sati ed famili theori capabl analog metalog framework base induct de nition background section provid background materi rewrit membership equat languag materi present standard postpon discuss ectiv aspect section rewrit logic rewrit logic simpl logic sentenc sequent form signatur theori rewrit logic tripl signatur set collect condit rule infer rule rewrit logic deriv rewrit theori logic point rewrit logic metalog framework rewrit logic framework logic formula formal element initi model equat theori infer system formal express infer rule rewrit rule rewrit understood modulo equat support exibl abstract kind infer equat care structur bookkeep formal sequent structur rule sequent rewrit modulo equat axiom rewrit theori underli equat theori rewrit logic choic equat logic attract choic term express membership equat logic logic overload function capabl express partial equat condit view equat theori rewrit theori obviou membership equat logic rewrit logic membership equat logic rewrit logic initi model provid basi reason induct membership equat logic membership equat logic express version equat logic full account syntax semant membership equat logic found de ne basic notion paper signatur membership equat logic tripl set signatur fs disjoint famili set call set sort kind pair call signatur function call element kind kind set model interpret subset carrier kind denot algebra ground algebra set variabl atom formula membership equat logic equat membership assert form term kind sentenc horn atom sentenc form xm equat membership variabl figur set membership equat axiom minim logic theorem sort formal sententi david manuel mesegu theorem formula formula formula formula formula formula formula formula formula theorem theorem theorem fig membership equat axiom minim logic theori membership equat logic pair nite set sentenc membership equat logic signatur partial express membership equat logic fact term sort term sort establish axiom correspond und ned error element emploi standard semant concept logic signatur fa collect appropri kind function interpret function symbol assign sort subset sort thought unari predic semant denot subset kind algebra valuat assign variabl kind valu satisfi equat overload notat identifi uniqu homomorph extens write denot satisfact hold note model function symbol alphabet unari predic fs satisfact relat extend horn formula atom formula standard write formula sati ed valuat model theori membership equat logic simpli horn theori view signatur sentenc languag write model set sentenc model theori membership equat logic initi model basi reason explain detail section write denot initi model membership equat logic theori model note note write object logic connect detail section rewrit logic metalog framework minim sort formula theorem theorem op formula formula formula var formula mb theorem mb theorem theorem theorem theorem fig modul minim restrict axiom horn emploi formula formal properti sati ed initi induct properti system system implement rewrit logic design explicit aim support execut speci cation ectiv comput theori speci ed function modul system modul function modul theori membership equat logic equat function modul assum execut rewrit engin rewrit techniqu oper semant develop system modul rewrit theori rule system modul necessarili termin semant function modul modul denot initi model membership equat logic rewrit theori speci ed syntax function modul form membership equat theori meet requir mention figur function modul symbol ascii charact improv readabl note syntax function modul syntact sugar introduc nite set membership axiom sugar version figur present membership equat theori equat membership axiom horn membership equat oper declar op correspond horn xn xn xn david manuel mesegu variabl kind declar reduc sentenc variabl kind kind explicitli de ned infer system determin di erent connect compon poset sort addit syntact write shorthand formula variabl kind formula sort induct section introduc concept plai kei rest paper de ne induct principl membership equat theori show theori introduc concept simpl set adequ illustr main idea carri applic induct principl membership equat theori membership equat logic subset equat horn logic shown immedi theori uniqu initi model induct principl reason properti respect model de nition sort de theori membership equat logic sort fc cn sentenc form kind formula free variabl sort signatur induct principl respect formula form rewrit logic metalog framework kind membership equat theori de ne induct schema logic signatur note conjunct true implic replac initi model membership equat sort interpret smallest set satisfi axiom equal interpret smallest congruenc satisfi axiom set interpret sort induct gener stage correspond herbrand model collect horn induct principl formal induct stage set induct de ned induct stage induct de nition sort amount induct proof ground term kind sort establish reason membership equat theori augment sound theorem membership equat theori membership equat theori minim logic figur de nition rise induct principl sort formula formula formula formula formula formula formula axiom formal induct structur proof minim logic note induct principl sentenc membership constitut induct de nition subset sentenc call gener sucient induct de nition addit proof oblig sucient complet chosen subset induct schema directli formal membership equat sentenc membership equat logic infer rule induct encod rewrit logic prove membership equat theori david manuel mesegu minim sort formula theorem theorem op formula op formula formula formula var formula mb theorem mb theorem mb theorem theorem theorem theorem fig modul membership equat theori carri metalog reason xed famili theori di erent wai theori membership equat logic paper notion extens theori describ extens parametr set axiom de nition theori membership equat logic extens paramet axiom membership equat theori tg signatur mutual disjoint consist constant call tg membership equat theori function tg denot instanc tg homomorph extens axiom substitut gener instanc membership equat theori axiom instanti instanc variabl replac ground term sg parametr axiom translat result signatur figur modul paramet kind sort set parametr axiom modul formal deduct theorem rewrit logic metalog framework rewrit logic logic framework logic point view rewrit logic framework logic infer system formal express infer rule rewrit rule rewrit theori note rewrit logic structur properti formula express nabl algebra data type satisfi equat rewrit deduct place modulo axiom neutral simplic rule rewrit logic ectiv appli logic framework exampl logic includ linear sequent present modal proposit horn logic lambda pure type distanc object logic rewrit logic virtual direct reason faithfulli simul reason origin logic wai conserv repres logic syntax infer rewrit logic mention simpl direct turn infer rule rewrit condit infer rule side condit underli membership equat logic repres logic sort membership equat theori condit membership axiom directli support rule typic present logic formal system modul present figur repres minim logic membership equat logic idea formula theorem minim logic term sort theorem minim note preserv induct natur set theorem proof minim logic repres famili logic sort membership equat theori theori figur paramet kind sort set parametr axiom theori repres famili logic includ extens minim logic axiom formula theorem minim logic extend sort membership equat theori repres famili logic correspond logic famili instanc membership equat logic correspond logic famili repres membership sort instanc membership equat logic david manuel mesegu axiom term sort theorem instanc abil repres famili logic import rewrit logic metalog give work report section ection rewrit logic section explain rewrit logic ectiv system implement ectiv rewrit logic deduct introduc boolean function ect metalevel membership relat membership equat section explain combin rewrit logic logic framework ectiv capabl build theorem prover carri induct proof ection rewrit logic rewrit logic ectiv theori function encod pair consist rewrit theori sentenc sentenc nite present rewrit theori term function de ned ht ti ht term section hold rewrit logic prove form ht ti ht ection languag design implement make systemat fact rewrit logic ectiv give user ned gatewai rewrit logic entri point prede ned modul speci cation modul term rei ed element data type term modul membership equat rei ed ground term data type modul modul complet de nition modul map theori term membership equat theori tg similar theori rewrit logic metalog framework paramet treat special parametr natur tg express tg fact paramet repres variabl sort term note de nition theori tg technic de ned ordinari membership equat theori extra tg ordinari ground term sort term variabl notat tg potenti depend tg entiti introduc chosen solv ambigu context occur introduc tg membership equat tg denot tg entiti rule appli term membership equat theori reason theori extens boolean function ect metalevel membership relat membership equat logic theori theori prove induct ground term sort function modul speci check membership base give full speci cation restrict subclass modul member modul membership equat theori axiom horn involv membership assert boolean function plai kei rest paper denot mention subclass modul membership equat theori tg write tg instanc tg tg tg write set variabl sort term paramet tg build induct theorem prover section present induct principl reason formula sort de ned function membership equat theori explain combin rewrit logic logic framework ectiv capabl build theorem prover carri induct proof paper detail build theorem prove tool david manuel mesegu build induct theorem rewrit logic infer system explain section ection de ne strategi control rule applic strategi infer rule speci ed rewrit rule necessarili termin import control applic rewrit rule order drive rewrit desir direct user control rewrit rewrit strategi detail de ning strategi induct theorem tool ectiv design function modul prove induct object level infer system induct proof data speci ed system modul metalevel encod syntax proof rule logic induct sort de nition introduc de nition di erent proof tactic guid applic rewrit rule infer rule de ned modul user submit initi goal pair form function modul sentenc signatur goal success transform infer rule rewrit di erent set case success subgoal left applic infer rule rewrit rule control user strategi note build theorem prover di erent level ection result modular design clean separ logic control compon simpli extend tool addit infer rule modul chang strategi languag de ned modul vice versa rewrit logic metalog framework show induct principl introduc section reason membership equat theori uniformli ect membership equat theori extens result present section provid basi rewrit logic metalog framework recent version tool rewrit logic metalog framework induct theorem versu induct induct principl present section prove properti ground term sort membership equat theori sort properti express formula signatur properti hold formula express hold initi model interest properti sati ed ground term sort express formula signatur fact typic prove induct de nition sort properti relat di erent membership equat theori properti natur formula signatur membership equat properti ground term sort ground term sort express formula notic boolean function express formula true situat similar prove properti ground term sort instanc tg theori tg gener statement tg extens fv vn kind paramet formal instanc tg tg ground term sort tg ground term sort express properti formula signatur instanc tg boolean function formal true vn tg true paramet sort claim instanc prove induct mirror expect induct proof show section crucial david manuel mesegu mirror induct reason sort theori extens tg induct reason sort de ned set horn fc cn de nition boolean function set ground term sort term true set term form ground term sort de ned induct set horn fc set horn fc cn idea fc deriv induct rule prove ground term sort obtain induct rule fc cn de nition mirror metalevel induct metareason mirror induct reason notic deal result induct rule quanti ed variabl repres paramet metalevel infer rule theori formal intuit introduc infer rule prove broad class membership equat theori correspond induct properti initi model modul gener form true tg true rang formula signatur modul formula instanc gener form case theori constitut trivial extens sound infer rule introduc base fact instanc tg tg set term sort term term sort tg induct de ned infer rule ect metalevel induct principl de ned de nition reason term sort instanc tg de ne extens tg sort set set horn induct de ne term sort tg paramet variabl sort de ne infer rule prove tg rewrit logic metalog framework de nition tg sort fc cn set sentenc form tg de ne fc tg true tg true tg true fx variabl kind sort term variabl replac variabl kind sort term note free variabl de nition tg fv sort fc cn sentenc form tg free variabl kind sort form true tg true paramet kind induct infer rule respect formula true form de ned de nition formula tg tg tg tg tg true sound infer rule prove david manuel mesegu build induct prover section ection design theorem prove tool explain ectiv design tool implement infer rule induct sort de nition carri formal extend tool induct infer rule extend version tool work report section section give illustr rewrit logic ectiv metalog framework standard deduct theorem deduct theorem minim logic present deduct theorem minim logic implic theorem interest reason begin central hold hilbert system logic justi es proof temporari assumpt manner natur deduct system illustr subtl aspect formal metareason deduct relat di erent deduct obtain rst ad axiom prove arbitrari standard statement deduct theorem statement relationship famili pair deduct system write denot theorem minim denot minim logic extend addit axiom belong result set theorem deduct theorem state formula minim proven induct structur deriv minim logic extend axiom formal minim logic rewrit logic provid minim figur extens introduc figur recal minim repres minim logic formula theorem minim logic rewrit logic metalog framework term sort theorem deduct theorem formula term sort theorem term sort theorem minim notic theorem state implic truth membership assert di erent membership equat theori famili formal deduct move claim formula formal deduct theorem initi model minim true minim true minim term denot variabl sort term object level preserv variabl follow convent term term includ element sort term perform metareason reason term respect membership equat theori minim explicitli assum formal deduct theorem standard textbook proof requir detail gloss trivial correct formal de nition boolean function fact minim conserv minim logic requir formula minim logic super drop proof provabl theorem extens minim logic axiom formula minim logic proof deduct theorem show prove note proof mirror standard proof deduct theorem prove appli ect version induct principl sort theorem extens instanc infer rule reduc prove prove formula figur notic conjunct correspond case involv prove deduct theorem induct proof theorem minim logic extend axiom rst formal case conjunct formal case instanc axiom schemata nal conjunct formal case prove instanc modu ponen david manuel mesegu minim true minim true true minim true true minim true true minim true minim minim fig goal result induct theorem constant membership equat logic reduc prove formula prove conjunct result replac variabl constant symbol sort assumpt minim proof result conjunct mirror proof case standard induct proof deduct theorem denot term sort modul result replac free variabl sort term constant symbol prove true true true minim rewrit logic metalog framework theorem constant reduc prove prove minim true assumpt constant sort term true true true note assumpt fact proven theorem extens minim logic axiom formula minim deriv formula minim true minim true minim true minim true prove equat proof mirror proof formula theorem minim proof mirror prove modu instanc axiom instanc axiom exampl experi rewrit logic ectiv metalog framework carri number proof formal base sophist version deduct theorem minim logic prove result similar basin matthew shown scope applic prove theori induct de nition present gener version deduct theorem appli extens languag axiom david manuel mesegu minim logic theorem deduct theorem hold minim logic implic proposit extens necessarili extens modal logic requir ad rewrit logic base di erent foundat consid object logic similar detail involv move level basic structur proof similar area appli result program transform ectiv declar point program transform program function act term properti satisfi understood paper ectiv declar methodolog program map program speci ed metalevel function add modul equat de ning desir object function structur induct sort de nition properti show induct rule nition comment experi prove theorem issu manag proof combin reason di erent level work logician comput ectiv metalog framework complic bit encod involv state carri proof reason involv level avoid practic problem work ectiv hierarchi exploit ectiv capabl build tool suitabl interfac hide level ection part built speci ed interact induct theorem prover section automat extract theori induct principl reason sort nition induct rule correspond ect induct principl metalevel task hand prove nition prove induct theorem amount comput strategi theorem case initi model interfac hide level note reason logic encod induct de nition logic framework isabel involv multipl framework theori induct de object logic addit languag write tactic rewrit logic metalog framework encod user user abstract detail focu essenti structur proof theorem abstractli approach reason base ectiv logic framework theori initi model initi experi demonstr machineri ectiv deduct membership equat logic rich foundat formal prove experi prove similar provabl logic framework base induct de exibl move theori prove theorem relat theori establish properti class theori requir pose build famili set induct de nition reason element realiz membership equat logic ection number direct work concern gener notion theori reason metalevel famili theori set constant make constant prove develop gener theori calculu reason metalevel famili theori arbitrari set axiom prove involv gener modul full illustr carri proof similar stronger framework logic interest formal comparison rel strength membership equat logic ection versu stronger metalog logic set theori relat question easi ect induct principl structur induct nabl order acknowledg research support darpa rome laboratori contract darpa nasa contract oc naval research contract nation scienc foundat grant author care read draft paper detail suggest improv exposit david manuel mesegu refer introduct induct de nition handbook mathemat page mesegu rewrit logic metalog framework technic septemb basin metalog framework logic page cambridg basin matthew scope intern workshop rewrit logic volum page electron note theoret comput scienc septemb basin matthew structur induct de nition inform mesegu speci cation proof membership equat logic theoret comput ection gener logic rewrit logic applic languag phd mesegu metalevel comput kirchner intern workshop rewrit logic volum electron note theoret comput page septemb elsevi speci cation program rewrit logic sri januari mesegu build equat prove tool ection rewrit logic symposium japan april formal wing formal volum lectur note comput page program proc mesegu principl intern workshop rewrit logic volum electron note theoret comput page asilomar septemb elsevi mesegu axiomat ectiv logic languag page san francisco april xerox parc mesegu ection strategi rewrit logic intern workshop rewrit logic volum electron note theoret comput page asilomar septemb elsevi mesegu ection condit rewrit logic manuscript submit primit higherord abstract syntax intern confer rewrit logic metalog framework type lambda calculi applic volum lectur note comput april ectiv modul algebra applic languag phd finitari induct present logic logic colloquium goguen mesegu model equal logic program volum lectur note comput page gordon introduct theorem prove environ higher order logic cambridg plotkin framework de ning logic januari mesegu rewrit logic logic semant framework technic report sri comput scienc august handbook philosoph kluwer academ publish mesegu gener logic logic framework logic page oxford basin experi framework theori logic page cambridg miller logic reason abstract syntax twelfth annual ieee symposium logic comput june mesegu gener logic logic colloquium page mesegu condit rewrit logic uni ed model concurr theoret comput mesegu membership algebra semant framework equat speci cation volum lectur note comput page mesegu research direct rewrit logic berger comput nato advanc studi juli august mesegu goguen induct comput algebra method page cambridg induct de nition system rule properti confer type lambda calculi volum lectur note comput approach implement de nition intern confer autom deduct volum lectur note arti cial june david manuel mesegu isabel gener theorem contribut tobia volum lectur note comput scienc autom theorem prove simpl metalog lf kirchner intern confer autom deduct volum lectur note comput page juli mesegu pure type system rewrit logic proc workshop logic framework mesegu approach connect applic proof translat sri februari van emden kowalski semant predic logic program languag 
AI/1594 2 decemb comput scienc ibm research report ai ibm research se june hong ibm research thoma watson research center yorktown ny research almaden austin beij watson tokyo zurich limit distribut report submit public ibm copyright accept public issu research report earli dissemin content view transfer copyright distribut ibm prior public limit peer commun specif request request fill reprint legal obtain copi articl payment copi request ibm watson research center box yorktown ny usa report internet ai ibm research apt morgenstern se june hong tj watson research center ibm research yorktown ny abstract ibm plai activ role ai research incept year ago trend reflect demand applic behav ibm todai carri ai research interdisciplinari fashion combin ai techniqu comput techniqu solv difficult technic problem introduct background contribut artifici intellig research date back arthur checker program made histori demonstr comput program plai checker beat human learn experi plai human player geometr theorem prover axiom geometri diagrammat inform prove basic theorem geometri earli dai ibm rich histori research broad spectrum ai research area activ ibm form ai group earli notabl contribut earli includ system queri natur languag comput algebra system subsequ commerci axiom system robot conceptu graph knowledg system develop pharmacolog knowledg expert system epistem logic nonmonoton reason ai research ibm base symbol paradigm research continu recent year trend statist ai applic area machin learn natur languag process led growth area statist learn theori bayesian network activ area inquiri report recent research tradit newer area ai section ai activ ibm research introduc extent ai ibm comput scienc research broad statist write reason research knowledg span wide varieti includ logic nonmonoton plan intellig agent discuss logic program nonmonoton reason ibm tradit logic program nonmonoton reason research watson lab forefront develop constraint logic variant logic program facilit arithmet reason gener reason comput system base constraint recent research logic program includ work benjamin logic program logic program specif scope potenti conflict mutual specif prioriti prioriti resolv conflict rule logic program form priorit logic type priorit default share high complex express form priorit default logic program implement common java librari enabl commun busi polici price promot retail custom applic includ xml format busi call busi rule markup languag pilot million industri consortium develop suppli chain integr manufactur default reason busi rule studi morgenstern singh develop network knowledg structur enabl effici reason potenti conflict busi rule fan extend tradit paradigm semant network inherit except attach logic formula node provid method determin formula node case method determin rule save discard involv inform inher topolog special rule give prefer class busi rule fan paradigm work busi straightforward interpret busi case situat node busi rule logic formula fan ibm develop expert system benefit inquiri medic insur system dynam configur bank insur product network scheme natur languag understand project ibm network structur innov wai improv perform difficult task text stori understand ibm research lab nobel gerd lead seek develop system understand mean natur languag research team depart standard natur languag process center anaphora word object identif system center world knowledg achiev slot grammar set transform rule gener textual input innov li transform repres understood textual input central part analysi fractal hierarch network arbitrari order organ input text structur isomorph exist fractal network isomorph facilit comparison text effort ibm research kr structur understand natur languag andrew index stori project seek organ larg knowledg base stori strategi motiv gordon develop languag repres strategi close model standard ai plan techniqu addit index languag applic interact entertain explicit reason strategi enrich interact experi erik work stori understand natur languag templat gener principl cognit behavior understand goal action system combin art natur languag process knowledg agent interact work extend earlier work fuse ai techniqu cognit psycholog intellig schedul research intellig agent develop core technolog facilit quick implement intellig explor foundat principl applic auction schedul build learn project led joe build hybrid intellig agent reason learn framework consist set core javabean set javabean agent construct compon editor standard environ project plan call build platform compat fipa intellig physic agent standard java intellig agent technolog form central part inform economi project head jeff studi role intellig agent set price varieti situat auction negoti situat vari wide term amount knowledg assum particip amount comput reason particip expect perform group studi strategi base set formal analyz behavior algorithm base strategi theoret analysi complement implement intellig agent program implement strategi retail consum kephart colleagu studi role reinforc inform inform bundl environ intellig agent technolog manag plan schedul problem richard colleagu examin difficult problem schedul multipl machin multipl object constraint problem examin schedul paper manufactur job factor batch size downstream schedul consequ consid simultan difficult problem achiev optim solut case minim total tardi machin group tackl problem develop agent architectur agent solut cooper exchang result algorithm implement basic result shown impress result achiev effici manner work receiv daniel wagner prize excel oper research practic contribut statist ai term statist ai denot area research methodolog ai statist brought achiev higher level intellig perform domain integr notic effect speech handwrit natur languag knowledg discoveri data mine relat applic machin learn speech handwrit recognit speech recognit technolog pioneer ibm earli largest group world dedic solv problem speech languag uniqu statist model approach long term research effort initi jelinek continu led highli success retail dictat product languag recognit project includ transcript radio broadcast relat activ includ speech biometr verif speech speech speaker recognit emerg thrust research activ combin speech recognit technolog natur languag process includ dialog natur languag statist spoken languag understand work lead system user speak natur tongu control applic person style emphasi scalabl dialog system capabl handl simultan input modal prototyp system applic rang stock mutual fund trade phone air travel shop languag includ spanish prefer handwrit clear pen paper suit task handwrit frequent conveni simpler current comput altern comput clear pen paper prefer task pen paper note frequent conveni simpler current comput altern pen technolog group tj watson research center believ tremend market pen comput system capabl satisfi high expect goal creat pen technolog group research aspect pen user interfac recent ongo algorithm research includ method perform handwritten document retriev improv handwrit recognit accuraci statist languag model natur languag process natur languag process ibm dynam research area span wide rang frequent overlap project human comput interact data mine commun pioneer algorithm system significantli understand field creat research direct accomplish includ earli machin translat system year air forc natur languag queri statist machin translat paradigm statist mathemat investig natur algorithm text categor base symbol rule induct statist theori pattern recognit work theoret issu comput linguist practic algorithm text machin translat knowledg manag algorithm develop colleagu extract inform enorm collect document web investig driven question answer user specif question mind valid answer differ level individu kei summari research lexic navig permit user find relat document connect origin document base set techniqu extract vocabulari item entiti domain relat link lexic network link concept amount inform present web comput explod abil read absorb remain essenti fix develop machin learn algorithm mine recur pattern linguist associ wealth data develop approach linear classif train quickli accur recent applic includ text categor email rout interact search web base concept dialog categor develop current algorithm automat topic discoveri newswir feed product aid analyz call automat identifi topic number call develop algorithm structur extract inform support automat updat unstructur text intent futur nlp system adapt domain mandatori manual enabl applic machin translat technolog crucial due wide usag internet exchang inform commun peopl world mt effort princip conduct research fulli automat machin translat algorithm explor method assist manual translat process linguist annot tool develop grammar check control languag check step make text machin translat develop extrem rapid statist machin translat algorithm specif tune inform retriev larg multilingu document collect effort advantag slot grammar base bilingu lexicon transform system produc accur machin translat broad coverag slot grammar parser european asian languag develop english slot grammar parser basi grammar check control languag check machin learn relat area ongo rapid growth data due internet widespread creat knowledg discoveri data mine methodolog challeng extract knowledg data draw princip research data pattern machin learn ibm research forefront excit area begin kei advanc robust scalabl data mine method fast pattern detect larg text web innov busi intellig applic worldwid research laboratori kei focu area kdd research ibm continu high perform scalabl data mine techniqu data repositori earli lead area establish quest colleagu associ rule sequenti pattern technolog effici detect pattern fast scalabl tree model invent technolog scalabl parallel data mine develop part project provid origin basi impetu data mine intellig miner principl focu area research kdd learn system predict data focu area research weiss rule base predict model integr data mine framework work result data mine middlewar rule base probabilist estim probe framework colleagu combin machin learn principl statist learn theori data manag technolog innov busi intellig applic area insur risk manag retail target market focu area applic machin learn perform manag manag kind comput web transact electron order good servic avail research ibm seek level autom perform manag work conduct earli includ expert system colleagu autom manag mv oper consol current work includ studi predict detect al advanc knowledg perform event mine form data mine recogn situat requir specif action ensur good perform intellig tutor system construct model understand interact student base model learn place group research scientist ibm expand paradigm interact team student develop intellig tutor system monitor manag team student collabor solv problem techniqu includ defin typolog collabor problemsolv store evid individu group perform bayesian network close relat area machin learn applic pattern recognit techniqu comput biologi solut discoveri repeat pattern amino acid sequenc gene function identif advanc pattern match algorithm develop bioinformat group activ pursu mine inform sequenc vision intellig vision requir abil understand imag camera sensori devic produc requir provid comput background comput inferenti method understand imag exploratori vision group ibm head studi vision system applic rang biometr video autom object recognit supermarket vari research reli innov technolog ibm research scientist develop method transform ordinari video camera segment function hand pose fingertip gestur imag iri research develop method recogn object partli occlud technolog develop includ color segment techniqu base simul method visual motion method face gestur innov sensor motion trajectori extrapol track motion techniqu foundat project ibm research biometr project explor wai secur system enhanc fingerprint recognit face recognit techniqu video analysi project explor video video function akin search string bodi harder perform high accuraci reason time constraint face recognit event recognit techniqu central project human comput interact aim aim emploi devic user commun aim touch specif hand gestur point grasp work intend partli peopl sever handicap troubl tradit devic face gestur recognit technolog central project project explor autom object recognit supermarket applic studi recognit veget regist problem difficult veget recogn tradit supermarket plastic partli occlud combin background subtract separ veget superflu inform concaten relev inform featur matcher find compar item memori regist number case close clerk correct match system learn mistak field test scanner manufactur program applic histori game plai began arthur samuel develop expert program learn experi forti year ibm research continu break ground develop program chess famou program deep blue made headlin beat gari program defeat world champion develop program long challeng problem ai program combin evalu function determin good posit effect search techniqu search space state difficulti develop program li fact state space larg branch factor number state reach state sophist evalu chess posit difficult develop perform deep develop murrai joe jerri chess champion joel reli combin brute forc sophist knowledg chess fundament success ibm sp comput power processor node yield processor work tandem net result system capabl calcul million move deep understand chess strategi implement evalu consid materi chess king safeti measur board tempo measur relat aggress rel game seemingli simpler game develop program prove difficult random element introduc game roll dice make appar intellectu game make branch ration higher hundr solut reli brute forc comput power primari techniqu gerald tesauro develop program reli form reinforc learn tempor differ learn explor machin train neural network learn complex function function evalu function evalu game posit plan strategi reli sophist evalu function learn expert expert chang mind board probabilist natur game mean expert rare interest game tesauro believ program learn evalu function applic program program valu ibm research solv challeng ai lai foundat crucial applic develop technolog deep blue central ongo work deep comput institut seek integr power intens comput advanc analyt area applic area comput bioinformat pattern data optim neural network technolog develop enhanc softwar horizon machin develop ibm hope leverag extend horizon comput aspect ai techniqu conclud remark contrast earliest artifici intellig consid ai recogn central part comput scienc transform attitud due chang natur comput live comput comput devic ubiquit everydai applic easier user applic understand anticip behav intellig ai integr part inform revolut clear combin ai disciplin work successfulli solv challeng technolog problem reflect technolog academ set ai techniqu found larg proport comput scienc depart trend evid ibm ibm research todai larg motiv project gener organ problem technolog tradit academ disciplin work hard ensur commun scientist similar research interest ibm research past year organ interest commun close parallel tradit academ disciplin pic augment replac tradit project organ ai strong compon project pic organ ibm research ai technolog plai major role project comput scienc research ibm current pic close relat ai artifici intellig pic area knowledg discoveri data human comput natur languag perform model comput biologi inform pic internet addit websit reachabl previou includ speech handwrit recognit acknowledg ai research ibm work introduc provid inform paper refer swami mine associ rule set item larg acm sigmod confer manag page white probabilist estim base data mine discov insur risk ieee intellig lr pv de ps ma robust method featur model continu speech recogn proc int conf signal volum page tool improv document qualiti confer appli natur languag associ comput page agent build learn environ project jh object imag system patent pf sa vj ps statist approach machin translat comput campbel knowledg discoveri deep blue commun novemb cooper byrd lexic navig visual prompt queri refin diglib damerau transform queri answer system oper statist american journal comput davi ibm telephoni system applic volum page jy vardi reason knowledg mit fipa websit foundat intellig physic agent gelernt realiz geometri theorem prove machin intern confer inform page gordon plai chess improv interact entertain spring symposium ai interact kephart shopbot thoma sixteenth intern joint confer artifici page morgan jh system acm page octob hy chan declar approach busi rule logic program xml michael acm confer electron acm sj hong iter approach gener minim rule exampl ieee transact knowledg data jf touretzki skeptic theori inherit nonmonoton semant network artifici ibm deep comput home page ibm ibm jaffar jl constraint logic program confer principl program languag rs scientif comput system york numer algorithm group isbn isbn page ts jl hellerstein rish recogn transact perform manag jk real time expert system jk jh sj base consult market ai white biolog inspir comput fourteenth intern joint confer artifici intellig page morgan yeh cooper multiobject support paper industri ma jl hellerstein mine partial period event pattern unknown period intern confer data mahadevan connel automat program robot reinforc learn artifici ol scalabl express medic terminolog mc transform system page mc slot grammar morgenstern inherit appli nonmonoton techniqu problem industri artifici mueller human comput model stream thought sh mae instanti speech biometr modul natur languag implement telephoni icassp june perceptu interfac inform joint process audio visual inform interact intern confer spoken languag volum page mp perron sd connel mean cluster hidden markov model workshop frontier handwrit jf eh quantifi contribut languag model writer independ line handwrit recognit workshop frontier handwrit radev predict annot combinatori pattern discoveri biolog algorithm samuel studi machin learn game checker ibm journal research visual inform improv speaker recognit speech process confer santa august mehta scalabl parallel classifi data mine intern confer larg boll automat text extract video annot retriev intern confer pattern mk algebra support teamwork manag role collabor learn environ cooper work jf sowa principl semant network morgan stein morgenstern motiv action formal theori causal reason artifici similar measur writer cluster workshop frontier handwrit tesauro achiev plai neural sm weiss oper data improv predict comput pattern detect intern workshop distribut oper manag mc improv natur languag process linguist document annot cole workshop semant annot intellig page sm weiss lightweight rule induct intern confer machin learn zhang ol text categor base regular linear classif method inform retriev 
HCI/1455 4 visual interpret hand gestur review vladimir pavlov rajeev sharma thoma huang depart electr comput beckman institut advanc scienc technolog illinoi mathew il usa abstract hand gestur attract altern cumbersom interfac devic interact visual interpret hand gestur achiev natur desir hci survei literatur hand gestur recognit context role hci number approach hand gestur recognit grown recent year systemat analysi aspect gestur interact develop discuss complet model hand gestur possess spatial dynam properti human hand gestur accommod natur type class model emploi interpret hand gestur hci consid util model human reli human hand imag investig model paramet analysi featur impact interpret hand gestur present light natur desir hci hand model offer complet model hand gestur lack simplic comput effici highli feasibl model suggest method effect gestur interfac hci integr hand gestur natur mode commun provid potenti answer problem work put advanc comput vision understand interact produc effect natur hand gestur interfac introduct human societi live interact entiti environ daili live interact peopl object perform varieti action import comput machin element societi increasingli influenc aspect perform interact environ concept interact interact comput advanc common hci reli simpl mechan devic mice joystick tremend reduc effect natur interact limit evid emerg concept surround interact virtual realiti recent studi shown natur point virtual space index finger explor virtual object hand easier understand peopl virtual environ manipul object idea virtual collabor environ research jointli design test prototyp product realiti mean hci perform interact environ natur earli dai comput attempt make understand speech year interest introduc mean interact field hci mean includ class devic base spatial motion human hand gestur human hand gestur mean interact peopl rang simpl action point object move complex express feel commun exploit gestur hci provid mean interpret comput hci interpret gestur requir dynam static configur human arm bodi measur machin attempt solv problem result mechan devic directli measur hand arm joint angl spatial posit group repres devic devic complet fulfil import requir hci natur gestur interfac forc user carri load cabl connect devic comput hinder natur user interact comput control environ overcom limit devic approach hci recent year approach suggest set video camera comput vision techniqu interpret gestur result interfac result burst recent activ area work gestur hci recognit static hand gestur postur varieti directli gener object recognit util imag geometr hand skeleton model exampl dynam hand gestur easili disregard interpret light object track hand gestur dynam action interpret motion hand convei mean postur fusion dynam characterist gestur hci recent role deserv numer rang global hand motion independ fingertip justif rapid growth hci brought surfac systemat analysi aspect light natur task accomplish technic literatur paper attempt achiev goal paper organ section formul problem hand gestur mean hci discuss model method hand gestur hci definit taxonomi gestur suitabl hci framework section section establish unifi approach analysi recognit hand gestur choic featur analysi human imag influenc model gestur perform gestur interpret scheme stress import hand gestur briefli discuss section number applic light analysi recognit techniqu present conclud review discuss limit current approach visual interpret hand gestur number solut prospect research direct hand gestur hci hand gestur mode hci visual interpret movement carri tremend advantag techniqu requir mechan restrict user movement restrict weight discomfort mechan devic visual interpret carri burden complex implement numer approach appli problem visual interpret gestur section approach chosen implement focu aspect hand pose hand postur effect studi process hand gestur global structur interpret system establish global gestur interpret system system requir mathemat model gestur establish model pivot success function system devot section discuss gestur model issu model decid system gestur descript mathemat model gestur model paramet space class grammar recognit analysi paramet model input video figur block diagram gestur interpret system classic path model paramet comput analysi stage imag featur extract singl multipl video input stream select featur specif task gestur interpret crucial effect model paramet comput note section section analysi stage recognit block paramet classifi interpret light accept model rule adequ grammar grammar reflect intern syntax gestur command possibl interact gestur commun mode facial express natur interpret gestur measur point encompass variabl number class movement cover requir mileston remain fulli reach futur gestur model qualiti gestur interfac hci directli relat proper model hand gestur model hand gestur primarili intend applic hci context simpl model suffici model establish natur gestur interpret comput discuss question model hand gestur hci definit gestur hci hand gestur easili defin relat commun aspect human hand bodi movement webster defin gestur motion limb bodi mean movement bodi limb express psycholog social studi tend narrow broad definit relat man express social interact domain hci notion gestur comput control environ human hand perform task mimic natur hand commun function classic definit concern mention human hand call establish mean interact comput control virtual environ definit hand definit vector describ pose hand arm spatial posit environ time paramet space hand gestur repres trajectori paramet space suitabl defin interv note definit suggest possibl gestur spite note gestur perform natur environ type except includ manipul hand modal question construct gestur model paramet set defin gestur interv focu attent question gestur taxonomi lack clear definit gestur gener taxonomi gestur taxonomi suggest literatur deal psycholog aspect gestur vari author author kendon distinguish occur independ occur associ mcneill levi recogn group icon metaphor taxonomi hci recent develop quek adopt gener taxonomi proposit taxonomi gestur applic hci figur classifi movement major gestur unintent movement unintent movement movement convei gestur inform gestur commun manipul manipul gestur act object environ commun inher movement gestur unintent movement commun manipul act symbol referenti modal deictic mimet figur gestur taxonomi hci commun natur environ accompani speech commun gestur act symbol symbol gestur linguist role symbol referenti action circular motion index finger refer speech modal gestur wing hci context gestur commonli gestur repres static hand discuss section act gestur directli relat interpret movement movement classifi mimet imit deictic taxonomi gestur larg influenc paramet space gestur interv determin relat gestur taxonomi classif gestur dynam issu tempor model gestur human gestur dynam process issu tempor characterist gestur practic natur resolv problem tempor segment gestur unintent movement question determin gestur interv psycholog studi gestur provid fairli consist answer previou question kendon call interv establish phase make nucleu stroke retract prepar phase consist preparatori movement set hand motion rest posit nucleu gestur form enhanc dynam hand return rest posit reposit gestur phase except rule call relat rhythmic structur discuss guid process tempor discrimin gestur set rule develop lead tempor classif set rule suggest quek formul modifi version rule form proposit hci environ set rule determin tempor segment gestur interv consist retract hand pose stroke classifi path paramet space gestur confin spatial volum repetit hand movement gestur manipul gestur longer gestur interv length commun gestur tempor phase distinguish gener rapid chang posit exhibit slower hand motion proposit hold case gener gestur hci section complex gestur interpret stringent constraint tempor variabl hand gestur work gestur hci reduc gestur static hand pose spatial model gestur movement action descript involv spatial properti hci domain influenc kind applic gestur interfac intend applic requir simpl model static imag templat human hand tv set control requir sophist hand model rise model movement provid complet descript gestur proposit complet gestur model hci paramet belong paramet space construct fx posit hand arm segment joint fingertip proposit reli assumpt human hand arm thought articul object valid hci deform human hand skin convei addit inform interpret gestur hci model provid inform requir correct analysi hand gestur hci hindranc approach paramet space high paramet obtain paramet model comput vision techniqu prove extrem complex overcom obstacl major approach gestur model util figur model gestur hand arm model approach examin approach close subsect spatial gestur model hand joint angl palm posit imag imag geometri paramet imag motion paramet fingertip posit motion figur spatial gestur model model employ model gestur model direct consequ proposit deal paramet mention proposit reduc set joint angl paramet segment length reduct accomplish set assumpt gener hold introduc depend joint bound move rang joint angl model base simplifi skeleton human research concern global motion cylindr model human arm bodi segment model human skeleton model common figur model mimic human hand skeleton kinemat exampl studi human hand morpholog biomechan found briefli describ basic notion relev discuss human hand skeleton consist divid bone bone bone joint connect bone natur exhibit freedom joint connect limit freedom movement hold joint figur finger joint show tm joint dof pip dip joint dof equal import notion dof notion depend movement neighbor joint natur peopl bend finger pip dip joint rang angl hand joint natur assum set constraint joint angl static dynam set constraint dof hand middl distal proxim distal proxim distal proxim middl ring thumb index figur model human hand static constraint finger thumb dynam constraint ip dip ip ip tm tm superscript denot movement joint center coordin system lee develop freedom hand skeleton model analog set constraint similar model equal lesser complex author model group model base imag mean model paramet encompass paramet mention proposit directli deriv model gestur relat gestur set templat gestur larg varieti model belong group base deform templat human hand arm deform templat set point outlin object interpol node object outlin approxim simplest interpol function linear function templat set variabl paramet describ variabl element obtain princip compon analysi train set data model simpl gestur classif base multitud class templat group model hand imag sequenc gestur templat gestur set gestur model sequenc repres imag element correspond view hand arm common view paramet model imag featur deriv imag complet imag sequenc human hand motion templat se gestur imag finger emploi templat finger track applic major paramet deriv imag templat denot class paramet hand imag properti paramet contour imag imag mention paramet featur analysi gestur section contour direct model paramet simpl contour polar exampl contour emploi basi eigenspac analysi paramet imag moment easili calcul silhouett contour paramet zernik moment orient histogram group model fingertip posit paramet approach base assumpt posit fingertip human rel suffici differenti finit number gestur assumpt hold space note lee palm assum finger limit number dof model locat fingertip palm applic concern deictic gestur singl fingertip refer point hand bodi gestur analysi previou section discuss wai model gestur hci analysi stage estim paramet paramet gestur model base number low level featur extract imag human oper act hci environ paramet gestur model acquir multistag analysi mono multi camera video input sequenc imag step constitut featur model paramet comput featur figur analyz step imag featur extract imag featur extract hand local segment hand local segment hand model paramet comput model paramet local imag featur camera camera global imag global imag figur gestur analysi hci hand local segment local segment process hand arm extract rest imag complex task lower burden local segment analysi varieti restrict restrict restrict restrict imag restrict background commonli distinct background greatli simplifi segment task addit restrict user requir wear long dark simplifi local problem restrict imag extract hand background perform threshold imag directli restrict setup emploi color histogram analysi color analysi applic characterist histogram footprint hsv color human skin mention approach requir addit process step applic resort uniqu color glove marker comput point method easier implement tend reduc natur interact techniqu advantag motion analysi move produc movement segment hand static object featur extract low level imag featur depend model gestur model type featur emploi calcul paramet similar model model finger trajectori requir fingertip extract featur imag featur wide scope paramet section featur obtain imag silhouett simplest wide silhouett easili extract local imag restrict background setup case complex techniqu emploi color histogram analysi hand exampl silhouett featur found hand analysi techniqu higher level featur paramet extract silhouett imag contour repres group featur edg detect scheme produc contour base simpl silhouett color imag contour model model analysi gestur recognit form set finger link candid model paramet instanc function point contour commonli featur gestur analysi fingertip fingertip locat obtain paramet hand model model detect fingertip locat space trivial effect solut detect problem mark glove color marker design characterist fingertip extract fingertip locat fairli simplifi perform color techniqu detect fingertip pattern match templat imag fingertip finger gener cylindr model techniqu enhanc addit imag contour fingertip extract algorithm base characterist properti fingertip imag curvatur fingertip characterist pattern heurist fact finger repres foremost point hand deictic indirect approach detect fingertip emploi imag analysi special tune gabor kernel discuss featur reflect featur combin featur form robust set effect paramet comput paramet comput comput model paramet stage gestur analysi gestur recognit stage recognit block track paramet comput stage produc final output type comput depend model paramet featur gestur model emploi success approxim method paramet comput basic idea vari model paramet featur extract model match obtain data imag figur match procedur begin palm end match finger initi model paramet select match gener hand posit obtain predict analysi paramet previou imag sequenc multitud featur hand paramet huang hand silhouett vari volumetr model paramet model silhouett match hand imag applic reli fingertip locat calcul model paramet lee prove locat fingertip addit characterist point palm uniqu defin hand pose assumpt similar discuss section model similar approach deriv kang al approach contour edg guid success adjust model paramet select candid finger arm link al al davi model paramet adjust model featur extract featur comparison model featur imag featur model paramet initi model figur model paramet comput success approxim techniqu deform model similar success approxim model orient princip compon adjust success satisfactori match model imag achiev simpler model direct map featur paramet space map explicitli defin moment imag emploi interpol correspond tabl obtain train gestur recognit gestur recognit phase trajectori paramet space analysi classifi member subset paramet space problem recognit optim partit paramet space implement recognit procedur problem partit design partit paramet space reflect percept determin class membership map paramet space relat implement issu recognit process question practic intens recognit procedur discuss question detail partit gestur paramet space influenc primarili kind applic hci system intend optim partit produc singl class paramet space gestur minim intersect gestur class provid complet descript gener partit reflect taxonomi movement tempor properti gestur tempor properti present proposit suggest tempor partit obei phase model model paramet space partit base tempor phase model initi final phase gestur inform follow gener taxonomi gestur proposit provid main taxonom differenti mimet gestur complet gestur model proposit induc paramet space partit rule produc set distinct class natur gestur hand gestur model closest complet gestur model provid possibl gener gestur recognit complet model wider class gestur cover system model fall short gener gestur recognit paramet space model insuffici describ gener gestur gener gestur difficult distinguish assum restrict restrict taxonom group paramet space optim partit respect gestur disregard dynam properti gestur analyz static postur restrict simpl model exhibit satisfactori recognit perform perform actual method time partit requir global motion distinguish tempor phase proposit model paramet space partit perform number classif method method requir repres gestur class class repres ad hoc determin hidden markov neural network membership map class base minimum distanc measur class repres hidden markov model techniqu case state hmm easili tempor gestur phase gestur hmm state hmm train procedur built base classif recognit procedur dynam time warp tempor invari classif gestur model hmm emploi recognit notabl success success recognit scheme gener base classif membership function context specif gestur establish introduc grammat element recognit procedur grammar reflect linguist charact commun gestur spatial charact manipul gestur subclass gestur action respect current previou state hci environ plausibl user reach valid manipul cup handl handl visibl user point hci system discard gestur small number system exploit fact grammar simpl introduc artifici linguist build learn user question comput effect recognit model complex versu recognit applic versu recognit time complex model wider class gestur appli comput complex recognit time gestur model ten paramet paramet calcul requir success approxim procedur price lower system base model rare show close perform time perform rang minut singl frame predict frame applic system gener hci gestur recognit arena superior simpl model model restrict applic narrow subclass hci enhanc comput concept hand postur classif instanc lower complex afford easier implement real time applic applic system recent interest gestur interfac hci driven vast number potenti applic hand gestur mode hci simpli enhanc interact desktop comput applic replac comput similar devic replac joystick button control machineri physic impair commun easili major develop gestur interfac growth applic situat virtual environ manipul gestur commun gestur gestur interfac manipul commun natur figur applic gestur interfac hci unlik gestur natur manipul commun gestur hci emploi direct manipul object convei messag hand gestur natur environ manipul action commun section commun role gestur subtl hand gestur tend support element speech except deictic plai major role human manipul aspect gestur prevail current applic hand gestur portrai manipul virtual object depict figur vo comput gener simul object window abstract physic devic control panel robot arm perform manipul object hci combin track commun gestur direct comput rotat object user interfac issu select object rotat object action hand track move pointer ve vicin object rotat object user rotat hand back produc metaphor rotat manipul pose commun gestur manipul commun gestur impli finit vocabulari gestur manipul natur movement answer question complex analysi recognit type gestur model section hand gestur model suit model manipul commun model gestur applic commun hand gestur model model section achiev usabl perform resort desir model gestur summari characterist system aim applic tabl applic hand gestur hci meant yield manipul action gestur hci convei messag storag transmiss process american sign languag provid opportun applic reduct bandwidth major issu typic solut problem code techniqu techniqu imag sequenc state physic object scene particip case updat descriptor receiv end comput gener model physic object driven receiv data code requir human bodi model appropri depend amount detail achiev model upper bodi limb fine tune model human face hand model gestur substanti applic applic gestur model techniqu gestur command complex cd player control panel hand silhouett moment track fp virtual squash hand silhouett moment contour track metaphor fp fingertip templat track na aliv templat correl track combin recognit facial express tv displai control templat correl track fp heurist detect point action track metaphor combin speech window manag hand pose recognit neural network track metaphor imag moment fingertip posit track metaphor fp heurist detect point action track dof hand model track fp silhouett zernik moment metaphor fp automat robot instruct fingertip posit grasp track na robot manipul control fingertip posit metaphor hand sign recognit discrimin featur imag sign na asl recognit silhouett moment grammar word fp frame tabl system emploi hand gestur hci speed measur complex interpret lack accur measur applic implement comput system level optim recognit asl consid applic natur emploi human gestur mean commun applic plai vital role commun peopl commun impair deaf devic automat translat asl hand gestur speech signal undoubtedli posit impact individu practic reason asl testb present hand gestur recognit system compar natur gestur human fact impli model techniqu suit asl proven recent applic prospect hand gestur hci vast applic mention section step introduc hand gestur hci develop natur point import issu excit applic hand gestur hci detail section futur direct fulli exploit potenti gestur hci environ class recogniz gestur broad gestur perform user unambigu natur interfac state art gestur recognit provid satisfactori solut achiev goal hci system present time address narrow group symbol command base hand postur type point section reason complex analysi recognit gestur simpl gestur model result gestur point direct quickli found silhouett human hand environ figur find hand distinguish simpl imag difficult interfac system base hand figur silhouett imag hand postur silhouett interpret reflect vertic ax silhouett silhouett unambigu defin hand postur base gestur model model commonli emploi hand track hand postur analysi analysi paramet hand model result wider class hand gestur identifi analysi link model lead point natur hand gestur model offer prospect model prospect present hinder lack speed restrict background hand approach issu adequ solv solut problem problem complex model featur extract fingertip posit featur section difficult extract solut problem emploi skin nail textur distinguish tip finger comput complex model paramet reduc optim number paramet satisfi level natur emploi parallel comput involv aspect pertain construct natur hci address futur aspect involv gestur human gestur natur emploi action hand gestur system focu attent gestur except system develop krueger inevit present time analysi techniqu requir hand extract global imag gestur ambigu situat occur case occur dealt distinct index versatil gestur analysi techniqu exhibit major speed techniqu handl gestur applic restrict simpl gestur requir hand adequ address issu gestur effect analysi techniqu consid techniqu reli improv classic techniqu exploit interdepend hand perform gestur case hand perform singl gestur assum symmetr postur issu relat gestur multipl gestur success interact environ multipl user virtual model task benefit enorm design simultan particip process implement interfac difficult issu foremost analysi gestur analysi present assum workspac gestur proposit case multipl user intersect workspac probabl event differenti user pose problem activ comput vision camera adapt focu area offer solut problem address issu interact multipl commun mode hci relat hand gestur hand gestur bodi mean commun section natur commun human concurr involv mode commun accompani gestur accompani word sentenc control deictic gestur involv index finger point control panel gaze direct panel commun gestur affirm complement mean speech messag literatur report psycholog studi human interact speech gestur mean commun explor lead multimod interact render hci figur affirm hand gestur gestur speech virtual environ gaze figur multimod hci system courtesi rich illinoi state reduc uncertainti speech recognit provid robust interfac gestur complement carri complet commun messag interpret speech gaze multimod messag reduc complex natur interfac hci design complic gestur command object select consist deictic gestur symbol gestur symbol object point hand simpl concurr deictic gestur verbal command number studi explor multimod hci steadili past coupl year present integr commun mode system perform command portion mode independ recogn interfac structur simplifi inform pertain interact mode lower level lost util multimod interact approach fuse multimod input analysi recognit consid futur studi interact human machin attract princip interest research past year coincid burst activ surround applic situat abstract natur surround virtual environ wide interact devic lower natur interact hinder effect environ direct natur mean interact hand proven plai essenti role solut problem paper focu discuss hand gestur visual interpret hand gestur yield potenti solut problem interpret hci comput control environ number approach hand gestur recognit grown tremend recent year systemat analysi aspect gestur interact emerg paper address task present unifi approach analysi recognit hand gestur visual interpret effect model hand gestur hci account characterist natur hand gestur complet model hand gestur reflect spatial dynam properti human hand gestur accommod natur type class model emploi interpret hand reli model human util human hand imag investig model paramet analysi featur influenc recognit hand gestur present light natur prefer hci hand model offer complet model hand gestur lack simplic comput effici highli desir accomplish model studi motiv potenti applic hand gestur natur interfac applic hand gestur interact system todai infanc find current system emploi hand gestur manipul complex interpret gestur dictat achiev gestur convei manipul action todai commun type hand gestur hci restrict produc singl user system downgrad effect interact suggest method elev effect gestur interfac show integr hand gestur gaze natur relat mode commun provid attract potenti solut problem substanti research effort connect advanc comput vision basic studi interact futur develop effect natur hand gestur interfac acknowledg work support part nation scienc foundat grant part grant electr industri refer hauptmann speech graphic intern journal vol pp feb prototyp mechan system concurr concurr tool technolog mechan system vol pp lu system workbench integr facilit intern journal intellig cooper inform remot control object commun vol pp fel neural network interfac speech ieee transact neural vol pp jan survei ieee comput graphic vol pp jan recognit ieee nation aerospac electron vol wang virtual point system robot inspect surfac flaw neural network base skeleton ieee intern confer robot vol pp issu studi biolog foundat motor semiot aspect pp lawrenc erlbaum mcneill languag activ studi deixi relat topic imag vision vol august hand gestur virtual realiti softwar technolog pp aug freeman control hand proc pp june base hand model intern confer comput june face gestur recognit video proc pp june analysi sign languag progress imag analysi process intern confer imag analysi pp descript gener cylind complex occlud system comput vol pp track recognit human proc pp june vol perspect vol pp oct hand model gestur recognit human comput master illinoi lee hand model techniqu comput pp lee analysi hand ieee comput graphic pp septemb rehg human hand tech rep school comput carnegi mellon rehg track articul tech rep carnegi mellon school comput pa decemb usabl hand ieee asilomar base hand pose proc pp june meyer recogn postur perspect pixel artifici neural network alexand elsevi scienc publish shape model train comput vision imag vol pp januari structur deform mode object long imag intern workshop automat june interpret human face hand gestur flexibl proc pp june darrel comput vision pattern recognit darrel express gestur analysi interact proc pp june input devic augment proc pp june cho shape ieee transact pattern analysi machin vol pp sept comput virtual realiti april interact freedom monocular imag proc pp june cui hand sign proc pp june moghaddam likelihood detect face proc pp june identif gestur input hidden markov ieee workshop applic comput pp decemb starner recognit american sign languag hidden markov proc pp june base hand gestur interpret asilomar confer freeman histogram hand gestur intern workshop automat june miss uncertain intern confer neural vol pp davi tech rep depart comput central human comput interfac user center pp point interfac imag comput vol pp point proc pp june hand gestur recognit window system proc pp june structur motion motion intern confer comput pp gestur input devic virtual ieee annual virtual realiti intern pp make real world commun vol pp juli ura gestur recognit edg proc pp june davi gestur iee proc vi imag signal vol pp april comput graphic driven verbal previou current activ comput vol pp davi hand asilomar confer wai oper proc pp june kang automat robot instruct percept recogn grasp ieee transact robot vol pp aug estim gestur input intern workshop automat june ieee vol pp artifici realiti ii time hand gestur recognit predict intern confer pp octob recognit human gestur imag process recognit gestur move ieee intern workshop robot human pp video commun system transmiss sign languag ordinari telephon imag sequenc process dynam scene analysi pp berlin imag code commun realist sensat recent ieic vol pp june blake activ vision mit vision visual ieee workshop visual applic open levi pp commun coordin gaze commun vol pp decemb multimod framework interact virtual proc symp human interact complex north septemb vo combin gestur speech adjunct april learn arpa spoken languag technolog workshop januari human ieic transact inform vol pp june 
ML/1198 6 simul evolut pattern recognit evolv hardwar tool build million neuron artifici brain hugo de gari michael donald coolei st michel europ tel fax suit boulder usa comput scienc utah state usa tel fax abstract paper present simul result evolut visual pattern recogn implement shortli real fpga base piec evolv hardwar implement genet algorithm evolv cellular automata base neural network circuit approxim complet run circuit growth perform evalu evolv download larg ram interconnect artifici brain architectur artifici brain million updat cbm rate billion ca cell speed enabl real time control robot birth research field call artifici built starlab control behavior life size robot call robot pattern recogn visual section artifici brain paper present simul result evolv gener properti recogn introduct paper present simul result evolut visual pattern recogn implement shortli real fpga base piec hardwar evolv cellular automata base neural network circuit modul electron modul modul assembl larg ram space artifici brain architectur ram updat cbm fast billion ca cell real time control robot cbm built deliv author previou lab novemb present lab april cbm essenti tool brain time write run year author web site journal articl confer paper project download basic aim project state build artifici brain billion artifici neuron year actual figur maximum million billion figur wait construct gener machin readi year cbm make practic creation artifici defin assemblag ten thousand higher evolv neural net modul defin artifici brain architectur artifici brain consist larg ram memori individu ca modul download interconnect architectur human ba ca cell ram updat cbm fast real time control robot neural net model fit evolv electron signal state neural net chosen bit binari label model collect bit neural signal present paper implement cbm paper report simul result evolv generaliz visual pattern detector modul base codi model ii input pattern recognit modul section describ evolut modul recogn static analog input pattern approach equal valid dynam pattern detect topic futur earlier paper convert back real time depend format codi modul cube cellular automata cell neural net grow evolv electron implement machin maximum input point maximum output output point evolv modul modul evolv input output point distribut evenli face cube ca due signal opposit connect input face problem confront map greyscal imag pixel input face cube map fairli straightforward shown tabl symbol form xi repres block pixel rectangl map face cube tabl map imag input face codi cube bought plastic plane digit photo side angl total photo imag input evolut train aim evolv modul give high output singl low valu plane imag imag test evolv modul capabl give high output signal low output signal imag train set digit imag transfer comput master convert grei scale pixel imag process softwar pixel grei scale imag initi convert analog input valu rang black pixel white pixel input pixel convert bit stream analog algorithm socal output spike train convert back time vari analog signal analog algorithm socal convolut target output tank set plane car fit function sum absolut differ target actual output analog clock tick tick fig show origin imag car photo angl pixel greyscal imag input evolv modul shown fig plane tank imag shown reason space paper fig origin test orient fig pixel test orient remain figur show elit modul evolv train set gener test set fig show output face imag train set fig show output test input imag angl star line line solid line car happi evolv result feel fig show achiev fig evolv output face fig gener output refer hugo de felix michael atr billion neuron artifici brain project year progress report artifici life robot hugo de michael felix michael hough build artifici brain fpga base machin appli mathemat comput special issu artifici life artifici brain comput felix hugo de michael simplifi cellular automata base neuron model artifici evolut octob michael hugo de felix hitoshi cbm hardwar tool evolv neural net modul fraction run million neuron artifici brain real time john marco david max hitoshi rick genet program annual juli hugo de michael 
DB/1788 3 queri network directori jagadish lakshmanan concordia milo divesh srivastava vista paper number abstract hierarch structur directori recent growth store address book contact inform person network resourc network servic polici system provid mean manag scale conceptu uniti autonomi multipl directori server superior convent relat offer deploi system data model ad hoc sophist requir navig access develop core formal data model network sequenc effici comput queri languag express power directori data model natur repres rich form heterogen exhibit real world answer queri express queri languag exhibit kind heterogen present extern memori algorithm evalu queri pose directori queri prove effici algorithm term complex data model queri languag share flexibl util recent data time effect address specif network directori demonstr mean repres exampl contact author room park florham nj work introduct hierarch structur directori recent growth larg number commerci directori server implement store address book contact inform enabl wide varieti network applic corpor white page electron messag internet engin task forc recent standard popular lightweight directori access protocol model queri network directori access network directori servic network directori view highli distribut directori entri organ hierarch namespac access search function ldap basi directori enabl network initi repres profil network applic polici manag directori applic section den applic directori wai complex current gener directori enabl applic thesi paper larg current gener manag browser applic provid interact access ldap ldap queri languag inadequ gener den applic den applic directori entri access long sequenc queri decad research queri languag prove advantag declar demonstr applic directori entri leav task determin effici access directori entri queri evalu engin directori seek bridg gap directori queri requir den applic construct provid ldap queri make ffl present formal descript core scalabl network directori data section spirit ldap dn illustr directori data model natur repres rich form heterogen network directori applic ffl sequenc effici comput queri section retain core ldap philosophi incur low resourc requir languag sequenc illustr specif class queri den applic support current ldap standard answer queri exhibit kind heterogen directori instanc ffl compar express power comput complex queri languag evalu algorithm current ldap section central result queri languag exhibit strict hierarchi express queri written express languag evalu time complex linear size input queri written express languag evalu time complex proport log input size survei relat work section conclud discuss section directori enabl network directori enabl network repres profil network applic polici manag directori introduc motiv run actual den applic studi qualiti servic core internet servic provid corpor administr servic level agreement defin contract user network introduct network discrimin packet provid multipl servic level achiev polici servic categori emploi bandwidth alloc flow maximum number flow support polici defin desir behavior multipl princip profil action profil identifi object relev action specifi desir defin valu collect attribut focu chaudhuri al repres polici directori directori directori repositori group administr constraint access network resourc base administr criteria simpl packet sourc address polici profil specifi packet polici appli polici action specifi treatment packet identifi constitut action enforc network administr polici forc constitut total import ensur consid defin unambigu treatment packet polici conflict occur polici directori conflict action conflict resolv popul directori describ mechan polici case conflict prioriti attribut case conflict highest prioriti polici prioriti attribut unambigu specif polici total order except attribut resolv conflict polici region polici except polici polici appli packet match profil except attribut easi insert delet polici directori figur show fragment sampl directori figur defer reader perspect data model section directori queri queri issu polici enforc proxi encount packet flow network packet type ip packet bear tcp connect enforc entiti attribut packet address destin address protocol current profil match directori action polici match higher prioriti polici appli polici except prioriti appli packet action appli enforc entiti condit packet stream locat devic independ order reach vast major telephon caller network address termin closest subscrib current offic home car telephoni packet network project goal provid simpl capabl subscrib move termin mobil termin reachabl den briefli describ top directori requir directori top subscrib repres network directori directori entri subscrib profil full authent set priorit subscrib polici determin subscrib reach polici consist queri handl profil subscrib control access reach set call repres wai subscrib reach caller satisfi call typic termin devic server consist set attribut identifi network address termin capabl call subscrib profil creat time top servic subscrib polici creat modifi dynam figur show fragment sampl directori queri handl profil call figur defer reader perspect data model section directori queri order call call applic queri directori logic subscrib obtain call caller suppli logic type media includ capabl call caller provid time compat caller termin match queri handl profil subscrib polici queri set call subscrib highest prioriti polici match inform subscrib control privaci inform call top applic receiv call present call call base current network directori data model present formal descript core scalabl network directori data base ldap dn suitabl den applic revisit motiv illustr model data network directori data model directori schema assum disjoint infinit set attribut class set type type denot abbrevi loss assum attribut includ basic type string int assum exist complex type domain consist sequenc set pair explain valu special type kei identifi directori entri definit directori schema ae finit set class ae finit set function associ type ff function associ set attribut class class call set attribut schema notion class plai role similar relat relat class model kei differ stem decoupl attribut type attribut defin independ class occurr attribut multipl class share type directori instanc relat model relat singl uniform data structur repres model forest singl data structur call node forest directori entri entri distinguish inform form set pair intuit formal assum infinit set object call directori entri definit directori instanc directori schema ae finit set directori function class associ directori entri set class val function associ directori entri set pair st condit commerci directori netscap directori server provid type deal telephon binari distinguish string note pair attribut belong attribut multipl valu directori entri entri pair iff exist class st pair iff attribut attribut rs type class directori entri iff class belong valu rs attribut dn function associ directori entri sequenc set refer distinguish sequenc call rel distinguish denot distinguish satisfi dn kei directori distinguish induc hierarchi directori entri entri parent entri entri child entri entri ancestor entri exist set pair st entri descend hierarch relationship distinguish parent note resembl distinguish fulli qualifi file unix system attribut file distinguish file unix model flexibl arbitrari set pair distinguish children directori entri observ directori entri attribut dn directori entri serv directori entri refer discuss sequel directori entri basic unit hold inform directori data similar record relat object model differ due model flexibl directori data model exampl directori instanc distinguish present section hierarch directori namespac influenc dn directori entri uniqu distinguish set entri organ hierarch hierarch organ call directori inform forest hierarch directori namespac typic correspond administr portion reflect organiz boundari network oper larg busi portion namespac oper directori server part namespac similar domain system serv mainten namespac distribut provid rapid lookup namespac refer directori inform formal model forest extens obtain closur properti queri languag att domain research figur higher level network directori inform forest network directori maintain highli distribut directori server provid directori servic limit number directori inform forest basic mechan akin dn time registr domain primari secondari directori server identifi owner hierarch namespac root domain entri directori server provid directori servic domain split domain directori server subdomain network directori servic suppli highli distribut fashion motiv applic revisit motiv network higher level directori inform forest correspond dn domain host hierarchi figur illustr portion higher level attribut dc class domain netscap directori server dot line demarc domain subdomain boundari directori server distinguish directori entri box surround figur exampl singl pair write dn sequenc singleton set simpli write dn sequenc pair secondari directori server ensur unreach network necessarili cut network directori servic qualiti servic core figur depict sampl data support qualiti servic base schema descript type directori entri servic level administr depict entri entri repres distinct entri entri repres distinct network packet entri entri repres inform tempor applic entri entri repres distinct action polici dn appli data traffic ip sourc address match weekend packet match profil deni permiss subnet polici traffic profil separ entri dn dn entri depict figur valid period polici distinct directori entri dn entri depict figur start time end time entri follow format action polici separ directori entri dn polici polici shown figur lack space locat devic independ top applic store subscrib data queri handl profil custom figur depict sampl top data model network directori data model top subscrib root child directori entri dn root directori entri profil top shown figur class valu attribut uid top queri handl profil children entri top subscrib sampl entri depict jagadish weekend higher prioriti lower prioriti jagadish work hour lower prioriti call child entri entri directori sampl call entri depict figur jagadish work hour offic phone higher secretari offic phone lower voic messag mailbox lower prioriti voic messag mailbox call weekend kei differ natur data model qo applic top applic hierarch directori namespac qo applic partit base tional hierarch directori namespac top applic partit subscrib top subscrib portion hierarch directori top subscrib repres manag profil action person ideal similar person directori applic natur network directori data model directori natur store data motiv applic relat reason directori relat directori data model defin hierarch namespac entri enabl highli distribut manag entri directori server permit conceptu unifi view data directli support relat model subnet network defin profil polici reflect heterogen entiti directori data model repres manipul heterogen natur flexibl critic ensur autonomi directori ffl directori entri belong multipl entri belong set class entri valu attribut definit requir singl class union attribut definit user belong class user belong class class default schema netscap directori server relat relationship directori defin subclass inherit ffl entri belong set class attribut semant attribut class definit form heterogen queri handl profil top subscrib valu starttim queri handl profil valu queri handl profil attribut ffl directori entri multipl valu attribut heterogen directori entri set attribut network polici rule multipl valu attribut entri multipl valu attribut relat model rigid applic record relat model belong tabl attribut singl object data model belong object homogen directori data model share flexibl recent model data time effect address specif network directori applic specif hierarch typic context network critic discuss arbitrari dag cyclic data easili attribut referenc entri refer support queri explain section queri languag boolean oper atom queri atom queri consist base directori search atom similar atom ldap queri base distinguish entri rel filter evalu scope filter evalu base entri children base entri descend base entri choic atom filter depend set base directori data model atom filter base type string int exampl atom filter compar individu attribut integ valu test presenc attribut wildcard comparison string attribut directori entri satisfi atom filter pair entri satisfi filter defin mean directori entri satisfi atom filter denot repres atom filter def def int def string queri function map directori instanc directori schema instanc schema compon remain result set directori entri semant queri denot definit atom semant atom queri scope valu scope base def fr bg def fr child def fr descend requir atom queri evalu effici major atom queri consid support evalu indic integ trie suffix tree indic string filter boolean oper atom queri combin boolean set differ form complex grammar figur appendix boolean oper obviou semant observ atom filter combin boolean form complex ldap filter complex ldap queri singl singl atom queri complex queri scope defin ldap queri languag virtual materi differ base entri locat directori entri surnam jagadish formul queri formul singl ldap queri section applic pose separ ldap comput differ applic boolean express evalu straightforward list merg input boolean oper repres sort list jacobson al describ eleg algorithm task sort list result boolean queri comput linear complex scan input list sort write output list result boolean comput list merg sort reason obviou evalu sophist queri sort order lexicograph order string distinguish directori entri queri languag hierarchi oper queri express network richer standard ldap advantag hierarch organ directori entri limit exampl languag extend hierarch select oper richer mean exploit hierarch organ directori entri present exampl describ formal semant addit oper illustr exampl parent queri organiz unit directli entri organiz unit locat queri entri locat queri queri singl obtain desir binari hierarch select oper children queri return entri satisfi operand binari children child entri satisfi operand binari children similar exampl model unambigu locat organiz person list detail ancestor queri traffic profil network polici govern subnet hierarch organ directori entri root directori entri formul obviou queri retriev traffic profil repres directori network polici desir traffic profil root entri directori entri locat queri queri singl obtain desir binari hierarch select oper ancestor queri return directori entri satisfi operand binari ancestor ancestor directori entri satisfi operand binari ancestor oper path constrain ancestor hierarch organ directori entri depict figur illustr entri children entri children entri locat closest ancestor entri directori entri queri subnet traffic profil govern smtp traffic port queri descend locat ancestor entri queri return entri satisfi operand ternari descend descend entri satisfi operand descend provid entri descend entri descend entri satisfi operand descend hierarch select formal explain extend defin present figur grammar definit hierarch select defin semant hierarch select oper def fr child def fr parent def fr descend def fr ancestor def fr descend descend descend def fr ancestor ancestor ancestor evalu hierarch select oper straightforward comput hierarch select parent independ test entri operand output find entri quadrat sum size operand kei effici comput parent children algorithm conjunct sort operand lexicograph order dn algorithm effici cpu time present adapt algorithm improv complex algorithm oper obtain judici modif basic algorithm modifi algorithm present figur algorithm work reason algorithm present argument understand correct subsequ algorithm algorithm sort base lexicograph order dn dn parent entri prefix dn child entri phase entri number parent children initi stack entri fi point entri lexicograph merg repeat push top stack entri top stack ancestor parent parent push top stack associ valu entri list pop stack entri process stack phase list scan result output entri repeat output output entri figur effici comput parent children paper correct algorithm base observ adjac entri stack correspond interven pair entri merg list pair merg list adjac stack point entri push top ancestor merg list present stack entri remov stack descend merg remov stack output algorithm sort lexicograph order dn algorithm extend comput hierarch select oper ancestor path constrain hierarch select oper ancestor descend comput ancestor maintain count entri number lower stack entri belong list number higher stack entri belong list encount count maintain increment entri push pop stack result algorithm comput ancestor descend algorithm figur appendix present detail differ algorithm figur highlight bold comput ancestor descend kei track stack entri list propag count stack entri count maintain increment entri push pop stack result algorithm comput ancestor descend algorithm figur appendix present detail differ algorithm figur highlight bold care analysi algorithm show linear complex crux proof observ stack entri swap eventu memori multipl time stack repeatedli grow process perform depend block number directori entri disk page theorem algorithm correctli comput algorithm correctli comput algorithm correctli comput complex algorithm complex algorithm block factor queri languag aggreg select oper power import queri support problem identifi highest prioriti polici rule manag problem locat top subscrib queri handl profil natur queri den import role plai aggreg queri languag suggest desir express queri involv aggreg network directori introduc aggreg languag requir care standard approach relat queri languag aggreg comput incorpor aggreg comput directli network directori model potenti requir abil dynam creat directori associ newli comput valu attribut place entri hierarch namespac destroi simplic famili queri simpli select directori entri input directori instanc associ newli comput pair exist directori entri mix queri languag updat result undesir argu aggreg select view primit incorpor queri languag extend languag support aggreg select distinct wai aggreg comput select perform result simpl aggreg select oper support aggreg select extend hierarch select oper ad extra aggreg select filter operand perform aggreg comput select relationship directori entri operand set operand oper present exampl describ formal semant illustr exampl aggreg queri polici rule manag subnet polici valid express aggreg select filter appli directori entri satisfi aggreg term associ directori entri singl number valu attribut entri entri greater return aggreg queri top subscrib queri handl express aggreg select variant children oper directori entri satisfi operand children children entri satisfi queri operand children aggreg select condit top subscrib children entri return aggreg select formal defin grammar present figur appendix aggreg select filter arithmet condit aggreg aggreg attribut integ entri aggreg form aggreg form definit simpl aggreg select directori entri entri aggreg ea form result appli ea denot def denot multiset valu set directori entri aggreg form ea entri result appli denot def set directori entri aggreg form result appli denot def gg semant simpl aggreg select queri form aa aa aa aa def fr aa aa aa integ ea depend instanti aggreg denot integ comparison oper definit structur aggreg select directori entri set directori entri entri aggreg ea form result appli ea denot def def def gg set directori entri function map entri subset entri aggreg form ea entri result appli denot def set directori entri function aggreg form result appli denot def gg queri form directori entri set op set entri set ancestor defin wit set denot coupl choic case similar def fr parent op def fr descend descend descend op defin semant structur aggreg select queri form aa aa aa aa aa aa def fr aa aa aa integ ea depend instanti aggreg denot integ comparison oper note hierarch select oper special case structur aggreg select obtain set aggreg select condit evalu simpl aggreg select simpl aggreg select express form sort list directori evalu scan input list individu entri aggreg form comput entri aggreg directori entri list aggreg form increment techniqu similar ross al aggreg list scan list directori entri determin result compar entri aggreg possibl aggreg depend form theorem simpl aggreg select express form comput block factor evalu structur aggreg select algorithm readili extend incorpor structur aggreg select call result algorithm refer algorithm invok algorithm appropri illustr focu extend algorithm figur directori entri list total number ancestor descend list select directori entri base count algorithm comput entri aggreg check aggreg select condit techniqu increment comput entri aggreg directori entri valu entri aggreg entri entri stack easili gener comput entri aggreg aggreg aggreg function count averag aggreg comput fashion figur instanti gener procedur aggreg select filter identifi entri list list differ algorithm figur highlight bold theorem algorithm correctli comput op hierarch oper aggreg select filter complex algorithm block factor queri languag refer oper network directori data model permit attribut distinguish valu dn treat refer directori refer insid directori relationship directori entri easili model directori entri view relationship multipl directori multipl directori singl directori entri support extend queri languag symmetr oper match directori entri refer attribut list entri directori entri list entri directori entri refer list match oper addit grammar queri queri formal figur appendix queri polici rule govern packet match traffic profil govern smtp semant express understood directori entri satisfi queri operand vd result provid refer directori entri satisfi queri operand vd attribut operand vd entri queri extend locat entri specifi qo action highest prioriti polici rule smallest simpl aggreg select oper symmetr dv refer formal definit refer semant vd dv aggreg select def fr def fr aggreg directori entri associ set express form definit wit set function extend def fr op vd def fr op dv semant refer queri aggreg select definit evalu refer straightforward comput refer oper test independ entri operand output find entri operand result quadrat algorithm effici comput oper algorithm sort base lexicograph order dn dn parent entri prefix dn child entri aggreg select filter phase associ list explicitli creat sort order initi list pair lp entri repeat specifi valu attribut append lp entri sort lp base lexicograph order dn column phase entri number wit lp global maxima comput initi rp initi repeat rp rp associ entri list entri lp phase list scan result output entri repeat output entri figur effici comput base techniqu join semijoin comput relat illustr focu comput aggreg select filter algorithm figur identifi entri list refer entri step algorithm creat list pair lp associ entri list dn valu attribut pair list lp wit entri list list sort lexicograph order list lp sort pair list lp partit base associ pair lp entri desir aggreg perform singl scan sort aggreg select perform final scan aggreg select dv oper perform analog comput symmetr vd oper perform similarli theorem algorithm correctli comput maximum number valu attribut entri list complex algorithm block factor compar assess express power introduc queri languag motiv construct suggest captur earlier languag formal captur intuit theorem proof follow gener line argument omit ldap ldap queri languag defin paper commerci ldap protocol compon queri languag aspect studi theorem ldap ae ae ae ae recal extend oper relationship theorem dg express pg express pg express fa express write fo ong queri languag obtain ad oper seri claim show languag fa express power languag strictli fewer oper reason design simpler write binari oper ternari oper express parent argument includ directori lead evalu algorithm complex linear size input argument appli children oper complex queri evalu queri express evalu atom queri result entri sort lexicograph order dn oper queri tree previou result pipelin higher oper queri tree oper sort input comput sort output addit sort result intermedi oper comput queri result oper complex evalu algorithm put result obtain theorem theorem queri languag comput constant size main complex jlj jlj cumul size output atom jqj number node queri tree block factor leaf node queri tree queri involv atom select assum comput scan indic result cumul directori entri process queri tree oper produc output directori entri oper evalu linear complex constant size memori evalu queri tree topolog sort constant size jqj worst case complex jlj theorem queri languag comput constant size main time jlj jlj maximum number valu attribut entri distribut queri complex den queri issu core network element proxi distribut network typic directori server network data relev queri manag queri directori queri process perform local directori server discuss data relev queri manag multipl directori server network queri express evalu atom base dn manag directori server queri issu directori server manag base dn atom queri directori server locat effici mechan similar dn result atom queri ship origin queri directori comput queri result algorithm relat work network directori model hierarch inform remind reader earli hierarch data model led develop commerci im kei differ earli dbmss provid access languag data declar queri languag ldap hierarchi central focu studi type system descript logic hierarchi class hierarchi orthogon class hierarchi person work compani entri class algorithm effici hierarch structur relev literatur algorithm check presenc set edg path jacobson al present linear time algorithm comput element list element context search web file system jagadish al present linear time algorithm comput element list element context support person organiz distribut list ldap directori build work algorithm larger class queri directori data model directori data model share flexibl model websql recent model data lorel unql time effect address specif network directori applic specif hierarch typic context network critic distribut perform arbitrari dag cyclic data easili attribut referenc entri semistructur model defin hierarch typic defin node content set pair research recent schema graph schema data guid unari datalog schema schema definit languag descript logic formal differ kind restrict object vari simpl upper bound set compon graph schema arbitrari regular express view approach complementari bodi plan investig integr futur discuss ldap directori recent gain tremend popular larg number directori server implement compani critic sun tandem move tradit role server contact inform address ldap directori servic wide varieti applic current futur oper includ window nt ldap directori servic manag os resourc ldap directori enabl network promot major player includ ibm microsoft paper repres attempt formal data model sequenc queri languag popular formul desir includ closur properti permit queri obviou current commerci ldap system demonstr inabl current ldap standard express queri support applic emphasi network seri demonstr languag seri enabl express specif class den support current ldap standard shown express achiev unduli comput cost implement construct queri languag specif directori enabl applic investig util class network directori applic refer wiener lorel queri languag semistructur data journal digit architectur directori servic transport packet telephoni intern workshop network oper system support digit audio juli architectur telephoni packet network ieee journal select area mihaila applic web queri languag intern www santa beeri milo schema integr translat structur data intern confer resnick structur data model object acm sigmod confer manag page june suciu ad structur unstructur data intern confer page suciu queri languag optim techniqu unstructur data acm sigmod confer manag june lenzerini knowledg fifteenth nation confer artifici intellig cardelli wegner understand data polymorph comput directori schema servic level administr differenti servic integr servic network draft submit directori enabl network ad hoc work group cisco directori enabl network consen mendelzon hy queri visual system annual confer manag page consen mendelzon visual formal real life acm symposium principl apr forrest directori enabl network drive goldman widom enabl queri formul optim semistructur intern confer larg string ldap search filter request comment dec smith program applic lightweight directori access protocol macmillan technic smith ldap url format request comment dec ldap world implement survei suciu search hierarch structur directori set seventh intern confer inform knowledg manag nov vista flexibl list manag directori seventh intern confer inform knowledg manag nov mccreight suffix tree construct algorithm domain concept facil request comment motwani infer structur semistructur data workshop manag comput directori enabl network standard complex aggreg multipl granular intern confer extend page ullman principl system comput scienc kill lightweight directori access protocol attribut syntax definit request comment dec kill lightweight directori access protocol request comment dec lightweight directori access protocol string distinguish request comment dec directori fragment algorithm sort base lexicograph order dn dn parent entri prefix dn child entri phase entri number ancestor descend initi stack entri fi point entri lexicograph merg repeat push top stack entri top stack ancestor push top stack associ valu entri list pop stack rb entri top stack entri process stack phase list scan result output entri repeat output output entri figur effici comput ancestor descend algorithm sort base lexicograph order dn dn parent entri prefix dn child entri phase entri number ancestor descend interven ancestor descend entri initi stack entri fi point entri lexicograph merg repeat push top stack entri top stack ancestor push top stack associ valu entri list pop stack rb entri top stack entri process stack phase list scan result output entri repeat output output entri figur effici comput path constrain ancestor descend algorithm sort base lexicograph order dn dn parent entri prefix dn child entri aggreg select filter phase entri number ancestor descend global maxima comput initi stack entri fi point entri lexicograph merg repeat push top stack entri top stack ancestor push top stack associ valu entri list pop stack rb entri top stack entri process stack phase list scan result output entri repeat output output entri figur effici comput ancestor descend queri queri queri queri queri queri queri scope figur grammar queri languag queri queri queri queri queri queri queri queri queri queri queri queri queri queri queri figur grammar queri languag queri queri queri queri queri queri queri queri queri queri queri queri queri queri queri queri aggreg aggreg aggreg figur grammar queri languag queri queri queri queri queri figur grammar queri languag jagadish uid jag jagadish research secretari weekend figur directori entri top applic deni research figur directori entri support qualiti servic core network 
IR/489 5 search world wide web steve lawrenc lee coverag recenc major world wide web search engin yield result coverag engin significantli singl engin index coverag engin investig vari order combin result engin yield time document averag compar result engin analysi overlap pair engin estim lower bound size index web million page internet grown rapidli incept decemb anticip expand year amount scientif inform number electron journal internet continu journal internet world wide web repres advanc retriev dissemin scientif literatur advanc educ introduct search engin altavista excit hotbot lyco northern light web view searchabl encyclopedia access scientif literatur long dream scientist web search engin made larg grow bodi scientif literatur inform resourc access scientif inform retriev literatur domin directli widespread group scientist major search engin compani claim size web continu index close entir web grow web rapidli grow inform present difficulti tradit inform retriev technolog tradit inform retriev system design environ typic index static collect directli access document natur web bring import question central architectur search engin expand number regularli updat detect reloc inform answer question impact search methodolog search web futur web search technolog number comparison provid rel coverag inform web search engin test involv run set queri number engin report number result return engin result comparison limit search engin return document queri term behavior occur inform retriev technolog engin requir exact match excit infoseek engin return document relat document longer exist engin delet invalid document document exist chang longer queri term addit document relev prevent accur estim coverag engin basi report number result selberg etzioni present result base usag log metacrawl meta search servic substanti search engin result significantli repeat result inform limit present percentag document user follow origin search engin result limit number includ fact relev difficult determin view present order affect user relev judgment present result percentag uniqu document return coverag engin result suggest engin cover fraction made experi consid percentag uniqu document top document return engin search engin return document selberg etzioni distinguish engin cover fraction cover entir web return document document queri result rank differ engin produc statist coverag major web search estim size recenc search engin major search engin consid alphabet northern light common percept engin index roughli document index larg proport web compar number document return search analyz search engin queri perform nec research institut methodolog retriev entir list match document engin retriev individu document analysi number nec research independ nj usa institut advanc comput colleg md usa tabl estim coverag engin respect combin coverag queri perform decemb confid interv hotbot comparison note result specif queri perform queri made state engin time perform note result partli due index engin index ident word document engin typic maximum file size effect truncat overs result due index reflect coverag engin search engin coverag ci hotbot altavista northern light excit infoseek lyco scienc vol april straint entir list document match queri retriev search engin order queri includ studi constraint import mention order engin rank document vari engin queri result document engin document engin uniqu url resourc found determin engin index uniqu url index url return subset url document document engin list match attempt download full text url document download queri term count constraint import detail search engin return document queri term number import detail analysi duplic remov total number document return engin combin includ ident page url queri consid engin treat capit queri differ altavista return capit result capit individu page page time includ analysi fix maximum document queri engin combin remov queri return document includ document exact queri term count word document match queri term form word exist document order document count match queri constraint engin morpholog queri special charact common word engin treat special charact differ stop word hotbot altavista list altern page special page includ statist engin specif identifi altern northern light document part publicli index analyz queri satisfi constraint tabl queri perform decemb queri initi made nec normal work period month manual check result retriev engin correctli engin period chang format list document request page document number consist check detect temporari failur search engin estim size web basi analysi overlap engin number import consid search engin typic index document hidden search form exclud document authent requir expect true size web larger estim search engin start index document interest estim size web engin index refer accur estim size web base overlap engin assum engin sampl web independ page index search engin user regist page reason assum user regist page engin page index engin partial depend sourc statist depend sampl perform engin fact search engin typic index page link popular page estim size web combin engin smallest largest term coverag reason expect larger engin lower depend index page page user regist index popular fig coverag number search engin queri perform decemb result normal averag combin averag queri averag queri significantli document return number search engin fig order estim size index web total number document web exclud page consid search overlap pair engin analyz overlap engin assum engin sampl web quantiti number document return engin number document return engin estim fraction index cover engin size index web estim number page index engin time hotbot index million page rel coverag valu tabl estim number page index engin tabl estim size portion web index analysi overlap pair smallest largest term coverag analysi limit queri return document avoid difficulti pair engin estim deriv averag note estim smallest engin smaller actual combin coverag engin studi million conclud result statist depend sampl individu engin search engin index web ci lyco infoseek infoseek excit excit northern light northern light altavista altavista hotbot report scienc vol april page web expect estim size assumpt accur engin size estim size web overlap larger engin estim index web million page overlap largest engin express engin coverag estim term fraction index web individu engin northern estim size web vari significantli internet archiv estim million page item imag forrest research estim million page altavista chief technic loui estim web million page wire digit report web million page decemb tom mitchel extrapol size report literatur produc current estim million page basi exist estim significantli underestim size web investig percentag document report engin longer valid page move longer median ag document return engin investig provid inform recenc search engin experi run decemb percentag invalid link northern hotbot time includ comparison result similar experi perform august rank engin term percentag invalid link chang significantli analysi median ag document return engin show similar experi perform august result suggest index pattern engin vari significantli engin recent page engin factor involv tradeoff size updat number drawn experi coverag search engin investig vari order magnitud estim lower bound size index web million page engin index fraction total number document coverag engin significantli limit basi estim size index individu engin cover index web engin limit network disk comput combin item claim contrari combin result multipl engin significantli combin engin studi cover time web engin engin engin largest coverag hotbot altavista scientist search inform occur place web home page scientist inform specif paper duplic link refer coverag search engin simplest mean improv coverag web search engin combin result multipl meta search engin metacrawl altern combin inform sourc major search engin autom onlin search internet softbot transform queri goal plan algorithm gener sequenc action order satisfi goal planner extens knowledg inform sourc access success softbot locat home page individu studi shake al search home page locat home page locat home page hotbot greatli improv possibl improv search scientist creation search engin design index page import scientist refer note internet capsul html wizard internet domain www barri scienc fox wire growth june june salton introduct modern inform retriev selberg fourth intern world wide web decemb annual meet american societi inform scienc url normal remov suffix trail remov port design remov segment domain url directori depth greater account machin charact url tild search engin typic upper limit number document retriev limit altavista northern check ensur limit januari upsid sci pr digit hotbot search site largest web million page extend enhanc web power decemb person commun etzioni commun acm sixth intern world wide web santa april anonym review comment suggest regist unregist trademark properti respect owner april accept februari fig coverag engin respect estim size index web queri perform decemb percentag index web index major search engin lower commonli believ note reason expect true size index web larger estim statist depend remain largest engin scienc vol april 
ML/2626 6 initi iter refin cluster algorithm bradlei initi iter refin cluster algorithm fayyad cori paul bradlei microsoft research june technic report microsoft research microsoft corpor microsoft wa intern confer knowledg discoveri data mine pp aaai menlo park initi iter refin cluster algorithm bradlei initi iter refin cluster algorithm cori ps bradlei microsoft research wa usa abstract iter refin cluster algorithm converg numer local minima sensit initi condit present procedur comput refin start condit initi base effici techniqu estim mode distribut refin initi start condit lead converg local minima procedur applic wide class cluster algorithm discret continu data demonstr applic method expect maxim cluster algorithm show refin initi point lead improv solut refin run time lower time requir cluster full method scalabl coupl scalabl cluster algorithm address cluster data mine background cluster formul wai machin learn pattern recognit optim statist literatur fundament cluster problem group data item similar gener approach cluster view densiti estim problem assum addit observ variabl data unobserv variabl indic data item data assum arriv mixtur model mix label hidden mixtur model cluster assign probabl data point call mixtur weight method assum number cluster input cluster optim problem find paramet mixtur model paramet compon maxim likelihood data model probabl distribut cluster form em algorithm techniqu estim paramet gener case find local optim solut maxim likelihood data maximum copyright american associ artifici intellig reserv likelihood mixtur model paramet comput iter initi mixtur model produc current comput posterior probabl data assum current model model paramet base posterior produc current model suffici focu initi step initi condit step algorithm defin determinist map initi point solut em converg finit point paramet local maxim likelihood data model determinist map mean local optim solut sensit initi point choic assum model repres mixtur data point posterior probabl gener mixtur compon focu mixtur model individu compon densiti exp covari matrix prior work initi method cluster question plagu procedur choic start point good solut random method present address issu initi assum randomli chosen start point method initi mean run cluster problem mention variant consist entir data randomli perturb time method random initi case em discret data valu initi mean coordin ax determin select coordin method initi em includ hierarch agglom cluster found em discret data initi hac show improv random initi initi iter refin cluster algorithm bradlei refin initi condit address problem initi gener cluster limit present result em good method initi exist compar standard method randomli initi start point method appli start point provid solut cluster problem cluster model perform determin mode joint probabl densiti data place cluster centroid mode cluster approach estim densiti attempt find maxima estim densiti function densiti estim high dimens difficult bump hunt inspir procedur refin initi point point closer mode challeng perform refin effici basic heurist sever subsampl data natur bia sampl repres mode guard possibl point tail subsampl overcom problem estim fairli unstabl due element tail sampl figur show data drawn mixtur gaussian mean left full data small subsampl provid inform mode joint probabl densiti function point thought locat mode underli distribut estim fairli vari exhibit behavior worthi note good separ cluster achiev observ solut obtain cluster small subsampl provid good refin initi estim true data cluster cluster overcom problem noisi emploi secondari cluster multipl drawn cluster independ produc estim true cluster locat avoid solut emploi procedur perform solv problem group point group fashion figur show solut obtain cluster mean depict show point obtain bs cs ds fourth problem determin group refin algorithm refin algorithm initi small random subsampl cluster em proviso empti cluster termin initi center set cm cluster solut form set cm cm cluster initi cm produc solut fm refin initi point chosen fm minim distort set cm note secondari cluster cluster em reason goal cluster solut hard fashion solv problem procedur secondari cluster includ hierarch agglom cluster cluster cm smooth cm avoid solut outlier includ refin algorithm sp start small subsampl algorithm cm small random subsampl data cm cm cm cm fm fm fm fm fm fm cm fm likelihood argmax fm return defin function call refin simpli call classic algorithm initi start dataset number cluster return set estim centroid true solut solut trial solut trial solut trial solut trial figur multipl solut multipl sampl figur gaussian bump full sampl versu small subsampl initi iter refin cluster algorithm bradlei cluster multipl subsampl subsampl multipl sampl solut cluster solut select solut figur start point refin procedur cluster argument perform iter procedur classic em slight modif classic em cluster check membership cluster membership cluster small initi estim empti cluster centroid set data element likelihood current classic em call initi heurist motiv termin empti cluster reassign empti cluster point maxim likelihood step likelihood set estim cluster paramet mean data set comput likelihood data set model scalar measur fit set cluster dataset em algorithm termin solut local optim likelihood function refin process illustr diagram figur scalabl larg refin algorithm primarili intend work larg work small dataset data set irvin appli classic em algorithm start point feasibl option size effici accur initi critic cluster session data set dimens ten thousand million record hour dai present method scale cluster larg specif target fit ram show accur cluster achiev improv result sampl base approach scalabl cluster method benefit initi method work small sampl initi fast sampl size full dataset trial sampl run time complex time cluster full larg initi sampl neglig size data set cluster algorithm requir iter cluster time complex small subsampl typic requir significantli fewer iter cluster reason expect budget time user alloc refin simpli determin number subsampl refin process small proport refin time essenti larg desir properti refin algorithm easili scale larg memori requir hold small subsampl ram secondari cluster solut obtain subsampl held ram note assum obtain random sampl larg realiti challeng task record order random sampl scan entir scheme reservoir note environ data view exist physic tabl result queri involv sort case oper special order result result view assum gener experi synthet data synthet data creat dimens data sampl gaussian element vector true sampl uniform distribut element diagon covari matric sampl uniform distribut number data point sampl chosen time number model paramet gaussian evenli weight goal experi evalu close mean estim classic em true gaussian mean gener synthet data compar random start point chosen uniformli rang data refin start point refin method size random full dataset number refin subsampl size classic em comput solut full dataset initi point estim mean match true gaussian mean optim prior comput distanc estim mean true gaussian mean true gaussian mean mean estim em full dataset determin quantiti initi iter refin cluster algorithm bradlei solut comput classic em full dataset simpli quantiti divid averag distanc true gaussian mean estim em initi start point result figur summar result averag random initi point determin uniformli rang data note em solut comput refin initi point consist nearer true gaussian mean gener dataset em solut comput origin initi point left summar averag distanc true gaussian mean show data model low effect initi strong high expect sampl produc cost sampl result solut effect mode reduc result public data present comput result publicli dataset primarili interest larg hundr dimens ten thousand million record data set method exhibit greatest reason cluster session larg affair refin start condit insur time invest pai illustr larg publicli data set reuter servic demonstr refin procedur data set uci machin learn repositori found data set low small number record small number feasibl perform multipl restart effici sampl size small begin initi effect data set interest report gener experi detail experi data set illustr method advoc appli smaller data set refin procedur suit data refin algorithm oper small determin initi start point speed converg full data order magnitud total time cluster situat note cluster label correspond cluster assign em object maxim evalu result data easi synthet data truth dataset uci ml repositori evalu method irvin data set discuss gener comment irvin ml data repositori imag segment data set data set consist data element dimens instanc drawn randomli outdoor imag imag repres instanc random initi start point comput sampl uniformli rang data compar solut achiev classic em algorithm start random initi start initi point refin method measur case report data extract model result refin method log likelihood refin refin irvin dataset evalu refin procedur data set fisher data set low size major result interest cluster data set random initi point refin initi point led approxim equal gain entropi equal distort measur case random start point lead refin solut refin provid expect improv reuter inform retriev data set demonstr method real difficult cluster task reuter inform retriev data set reuter text classif deriv origin data set made publicli part reuter part reuter carnegi group david lewi data consist document document articl belong higher level categori hierarchi reuter consist word count document hundr thousand experi select frequent occur instanc dimens indic integ number time word occur document document classifi categori cluster reflect categori task find cluster reuter result data cluster entir requir larg amount chose evalu result randomli chosen start condit result shown detail data set initi iter refin cluster algorithm bradlei refin method log likelihood refin refin chart show log likelihood measur solut sampl normal result case simpli divid document belong categori measur qualiti achiev cluster measur gain inform categori cluster pure cluster qualiti cluster measur averag categori puriti cluster case averag inform gain cluster obtain refin start point time higher inform gain obtain refin initi point conclud remark fast effici algorithm refin initi start point gener class cluster algorithm present refin algorithm oper small subsampl requir small proport total memori store full make approach appeal cluster problem procedur motiv observ subsampl provid guidanc locat mode joint probabl densiti function assum gener data initi gener cluster algorithm true cluster found cluster algorithm iter fewer time prior converg import cluster method discuss requir full iter costli procedur set method abil obtain substanti refin randomli chosen start point due larg part abil avoid empti cluster problem plagu tradit em refin reset empti cluster point reiter em start point obtain refin method lead subsequ cluster algorithm solut intuit confirm empir result refin method present context em note method generaliz method initi algorithm gener discret data mean kei insight algorithm cluster cluster subsampl algorithm produc model model essenti paramet paramet continu space stage cluster cluster step algorithm refin section remain algorithm step reason goal stage find case harsh membership assign desir refer gaussian vol pp neural network pattern recognit oxford press street concav advanc neural inform process system petsch pp mit cluster algorithm larg proc intern conf knowledg discoveri data mine aaai cheeseman classif theori pp mit ap nm db likelihood incomplet data em journal royal statist seri ro duda pe pattern classif scene analysi john wilei son scienc commun acm bradlei initi point proc intern conf machin learn morgan kaufmann dimens dimen ion ref ref log dimen ion dimen ion ref ref figur compar perform initi iter refin cluster algorithm bradlei fisher acquisit increment conceptu machin introduct statist pattern san academ note sampl tape commun vol kaufman find group john wilei son comparison cluster microsoft research technic report wa inform retriev data structur frake pp prentic multivari densiti wilei gener converg theorem local ieee tran pattern analysi machin vol bw densiti estim statist data chapman mixtur bayesian microsoft research technic report wa 
DB/105 3 usenix annual technic san june abstract dynam function placement cluster comput david gregori gibson carnegi mellon optim partit applic filesystem function cluster client server difficult problem due dynam variat applic resourc workload mix paper present system monitor dynam function placement applic manipul larg data set exampl workload show import proper function placement depend dynam perform differ frequent reach evalu prototyp adapt system includ variat filter variat applic resourc experi adapt situat adapt converg quickli case perform impact introduct effect util cluster resourc remain difficult problem distribut applic high cost remot versu local perform larg number applic sensit distribut function network effect cluster research support darpa order issu indian head contract addit support provid member includ lsi seagat resourc requir load proper partit function produc consum softwar engin techniqu modular object abil partit applic set interact solid techniqu determin cluster function deploi system continu reli complex manual made programm system administr optim placement function cluster difficult answer optim function placement depend varieti cluster characterist commun bandwidth rel processor speed workload characterist byte move instruct execut basic hardwar characterist chang fail constant system characterist determin applic invoc depend input paramet worst chang due applic chang phase competit concurr applic share resourc system fit solut case perform focu import class applic cluster applic select manipul larg data set applic benefit abil spread parallel comput exploit comput resourc reduc requir network bandwidth effect function partit applic import process power reach devic applianc abund process cycl recent led research augment storag server support execut code refer storag applianc commod activ disk file server remot execut programm storag server addit observ applic characterist simplifi task involv dynam function placement applic move process amount enabl monitor system quickli learn import commun pattern resourc requir inform system rapidli identifi function move reduc commun overhead resourc content prototyp call function data stream move back client server dynam condit programm explicitli partit function data stream distinct conform intuit program model runtim system monitor resourc consumpt commun intern box measur model decid reloc compon optim locat describ design implement set experi evalu abil adapt chang condit explor adapt variat network applic cach access applic data reduct content share phase applic dynam competit resourc concurr applic preliminari result improv applic time select placement placement function initi start node outperform static placement situat dynam proper placement vari execut effect adapt function placement base black box remov programm burden function placement remaind paper organ section discuss relat prior work section describ design section discuss program model applic built section describ system section present varieti experi demonstr dynam function placement abil effect adapt dynam condit section summar contribut relat work exist larg base excel research practic experi relat code mobil cluster larg fulli section discuss relev previou work adapt function placement relat previou system demonstr function placement autom accur profil commun resourc consumpt system histori make good function placement complement previou system dynam adapt placement condit river system dynam adjust rate match product rate meet consumpt rate variat adjust adapt cluster system perform task complement river adapt function placement dynam presenc multipl task system dynam balanc servic singl client server build work develop mechan dynam function placement realist cluster includ complex resourc resourc workload variat hybrid ship techniqu dynam distribut queri process load client server manag system techniqu priori knowledg algorithm implement queri oper estim partit work client server appli wider class applic reli monitor make placement knowledg semant algorithm implement applic compon process migrat system develop mechan cluster explicitli support function placement adapt commun mobil program system emerald rover support mobil applic leav migrat applic programm mobil agent mole agent tcl enabl agent migrat provid algorithm mechan decid agent provid mechan autom migrat explor benefit dynam function design implement prototyp system port test applic consist program model system goal make program model easi applic programm simplifi task system migrat function monitor resourc consum runtim goal improv effect achiev low monitor overhead design scale larg cluster size prototyp larg meet goal program model encourag programm applic function independ compon object mobil object provid explicit method checkpoint restor state migrat applic filesystem repres graph commun mobil object graph thought root storag server anchor storag object client anchor consol object storag object provid persist consol object part applic remain node applic start consol part data intens serv interact user rest system start node typic consist main function program consol part initi invoc propag rest graph shown figur system consist migrat invoc bind manag resourc monitor manag resourc short compon creation refer mobil redirect method invoc face object enact object migrat bind manag notifi local resourc manag procedur call return mobil object resourc manag notif collect statist byte move object resourc object amount memori number instruct execut byte resourc manag monitor load local processor resourc manag updat stat bind manag migrat resourc manag client anchor object applic object migrat raid object bind manag storag server anchor object figur illustr object princip interact show filter applic access stripe file function partit object dark oval depict mobil clear oval mark anchor object method invoc transpar redirect locat transpar invoc compon runtim compon updat local resourc monitor compon procedur call return mobil object arrow label client period send digest collect inform server resourc manag server collect relev statist initi migrat label experienc stall time network transfer storag server activ access local mobil object resourc manag collect statist resourc manag emploi analyt model predict perform benefit move altern placement model account cost includ time wast wait object quiescent time wast checkpoint transfer restor target node analyt resourc manag arriv placement net benefit placement current object migrat place prototyp written leverag featur simplifi write mobil object commun rpc focu prototyp function placement address import orthogon mobil code issu issu code mobil dynam link requir migrat modul static process client server issu protect address import issu product tangenti question address paper program model program model princip mobil repres unit migrat iter process defin mobil object entir data process applic mobil object mobil object explicitli declar programm consist state method manipul state mobil object requir implement method enabl system creat instanc migrat mobil object medium meant simpl primit perform process step data pariti aggreg mobil object privat state access export interfac implement mobil object intern object opaqu mobil object system privat state consist object refer extern object mobil object save privat includ state method call reinstat includ creation initi system invok migrat node method save state buffer file method reinstat state locat method invok extern invoc activ mobil object storag server server data local storag object export flat file interfac storag object access server host migrat migrat portion applic li storag object side consol object applic declar object object implement log declar filesystem effect anchor storag server start server host iter process model synchron invoc start toplevel consol object propag object graph invoc return back consol object result number applic record process invoc return object graph longer activ object activ start invoc initi consol object initi invoc request receiv object occur object perform background work cach assum common mode oper number entir file data set iter properti requir monitor migrat system accumul statist return method invoc make object migrat program make singl procedur call stack collect valuabl inform end point migrat useless exampl stress program model evalu benefit adapt function implement distribut filesystem data intens applic describ section report perform section distribut filesystem applic requir varieti servic underli storag system enabl filesystem explicitli migrat provid storag servic reliabl function approach pioneer filesystem work spring oper system filesystem coher file directori abstract atop flat file space export base storag object file stack object creat repres servic bound file file includ raid object stack file object turn instanti lower level object object graph access file start object stack system propag access lower layer prototyp filesystem distribut addit layer typic found local filesystem cach servic support file directori share filesystem file data directori data cach manipul trust client multipl client concurr share implement af style callback cach coher multipl client concurr updat directori filesystem includ protocol ensur updat perform client consist commit server scheme highli scalabl absenc content requir lock server lock traffic section describ automat concurr control protocol high content lock scheme simpli adapt object placement consol app client cach raid raid isol atom storag consol app directori client network server directori cach raid dir dir isol isol atom atom storag applic interfac vf layer interfac directori block cach updat file cach raid race detect log persist storag typic object placement filesystem object figur architectur distribut filesystem built atop figur show typic file directori object stack object placement shown default network trust client shown compon filesystem object implement date descript function graph consist filesystem object provid vf interfac cach option raid storag object ensur pariti corrupt race involv concurr write raid object anchor storag server object intercept read write base storag object verifi consist updat commit protocol isol object highli scalabl cach object index block share cach filesystem process raid object stripe maintain pariti individu file set storag server storag object provid flat storag configur standard linux filesystem attach secur disk prototyp back store figur show sketch typic file directori stack graph consist directori object storag object directori object directori call cach directori entri object support cach coher optimist concurr ensur atom write directori perform implement object special directori semant raid object object ensur cach coher read write instal callback cach block read call break relev callback write ensur proper concurr control simultan updat timestamp cach block export special method check readset conflict atom write provid ensur set block differ object updat entireti updat log share instanc object filesystem access wai applic includ object directli append object subgraph applic object graph file open filesystem mount standard redirect unmodifi applic standard system call interact filesystem legaci applic mechan legaci applic benefit filesystem object adapt migrat beneath applic applic similarli object perform oper data mine port straightforward search applic iter input data invok success read call filesystem oper buffer time port readset consist list block read client directori oper requir read directori block ensur exist updat block insert number readset case directori block timestamp block updat kind applic simpli requir encapsul filter compon search object write method method state consist posit input output content current buffer system system consist bind manag resourc manag bind manag instanti mobil object invoc method manner migrat object cluster node resourc manag collect statist resourc usag avail measur adapt placement improv total applic time object instanti invoc kind node cluster client server server node base storag object resid client node execut applic access storag server server execut storag server potenti client server applic instanti mobil object make request system filter applic consol object request filter object creat system creat object memori invok oper alloc return caller uniqu call object caller rid invok mobil object rid act layer object refer object current locat bind manag mediat method invoc rid forward current rid gener concaten node identifi object creat local object identifi uniqu node locat maintain inform perform invoc hash tabl map rid object refer pair mobil object move tabl updat reflect node object refer node rid pass argument method system properli redirect method call web object constitut graph node repres object edg repres invoc object object address invoc implement procedur data pass extra copi object commun machin address remot procedur call emploi object migrat addit properli rout object bind manag enact migrat migrat object sourc node target node bind manag sourc node block call migrat object bind manag wait activ invoc migrat object drain migrat cancel step long object checkpoint local invok method object alloc buffer store state write filesystem checkpoint size larg state transfer restor storag node locat tabl sourc target node updat reflect locat invoc unblock redirect proper node updat locat tabl procedur extend migrat subgraph object locat tabl accur provid hint stale node sourc target cach hint locat updat migrat occur stale data detect correct node attempt invok object node node notifi object migrat scalabl node requir maintain forward pointer object host point past node inform caller current locat object node redirect caller node object call home node home node easili determin encod rid home node inform locat migrat locat tabl updat addit tabl sourc target node object move client home extra messag requir updat locat tabl migrat resourc monitor system intermediari role redirect call collect statist monitor code procedur call return mobil slow execut method mobil object section explain statist collect singl thread cross boundari multipl mobil object make method invoc propag stack resourc manag charg time thread spend comput block object charg alloc memori proper object collect requir statist previou refer observ window describ statist data flow graph byte move object monitor inspect argument procedur call return mobil object number byte transfer object record time data flow graph graph maintain move averag byte move pair commun object graph data flow graph tractabl size applic bulk process fashion small stack object memori consumpt monitor amount memori dynam alloc object procedur call return mobil pid thread make call record point system process mobil object activ thread address space wrapper memori alloc routin inspect pid thread invok memori alloc routin pid determin current object object charg memori alloc freed instruct execut byte number byte process comput amount monitor number instruct execut object observ window process rate amount measur time spent comput object combin linux interv timer pentium cycl counter track time spent process mobil object stall time estim amount time thread spend stall inform provid system timer extend system call support type denot block timer decrement thread block implement kernel updat real timer activ updat block timer thread queue mark block interrupt task dynam placement resourc manag server seek perform migrat result minim averag applic time applic access amount figur subset object execut client benefit comput closer data migrat object server potenti reduc amount stall time extend time object spend comput processor overload resourc manag server analyt model determin object migrat client server migrat back server client analyt model consid altern placement configur select net differ benefit move placement cost migrat net benefit repres estim reduct execut time migrat enact resourc manag find placement net benefit exce configur threshold avoid migrat chase small set reflect confid measur model system resourc manag commun figur global optim placement resourc manag decid altern placement applic stream access design robust scalabl reason detail comput requir estim net benefit discuss technic report outlin intuit comput resourc manag receiv measur receiv statist client processor speed current load collect similar measur local system local execut object data flow graph measur stall time latenc model estim chang stall time object locat instruct byte rel load speed estim chang execut time object placement addit chang execut time migrat model estim chang execut time object execut target node result load object treat memori server fix constraint stall time execut time amount benefit placement comput analyt model assum histori repeat window observ cost placement estim sum fix cost time wait object time transfer state sourc destin node estim size checkpoint buffer bandwidth node perform evalu show perform depend placement function subsect follow give increasingli difficult case adapt function placement correct locat hard imposs anticip includ scenario correct locat base hardwar applic applic data access content share data includ scenario stress adapt dynam phase applic behavior content multipl applic perform fair comparison applic run run network filesystem implement differ nf differ give applic advantag adapt function placement evalu environ evalu environ consist client storag server twelv node standard pc run redhat linux equip pentium ii processor main memori experi exhibit page activ server singl id disk drive averag media transfer environ consist switch refer san share refer lan storag server directli connect client connect san san client resid lan lan lan bridg san link network low perform rel speed similar san lan environ san bar graph section adher common format graph show time configur experi migrat object report object static locat begin dynam monitor system potenti migrat static storag server graph confid interv report averag run confid chosen smaller benchmark underscor abil adapt quickli note absolut benefit achiev dynam function placement function durat longer benchmark oper larger file amort adapt delai experi observ set threshold observ window adapt network issu network dictat rel import commun rel server load evalu abil adapt network topolog default execut function client offload contend server server client benefit server requisit cycl goal decid benefit execut due reduct network stall time exce slowdown due slower process raid isol isol atom storag client raid server san raid isol atom storag raid server client lan figur figur show file bound applic access raid object maintain pariti code access storag object run storag devic show bind stack raid object run bind raid object run storag devic thicker line repres data move configur depend bandwidth client storag devic client lan client partit lead lower access latenc experi softwar raid function move amount data touch byte xor content multipl file filesystem bound raid object storag stripe raid object maintain pariti stripe data multipl storag distribut concurr access share stripe client concurr control protocol raid object execut client node storag server object graph file experi shown figur proper placement raid object larg depend perform network connect client storag server recal raid small write involv data pariti write data pariti disk failur block read requir read block stripe reconstruct fail time lan san client adapt server degrad san figur figur show result raid benchmark content cpu resourc make raid lan network bottleneck data result substanti network traffic raid object storag server construct workload evalu raid perform consist client write separ file sequenti stripe size data stripe unit workload consist client read file back degrad mode disk mark result shown figur execut raid object server improv raid small write perform lan case factor execut object host perform experi adapt place object optim san execut raid object local client faster client lower load perform raid function quickli arriv advantag raid slightli pronounc degrad read optim locat fast server arriv optim automat select locat raid object adapt paramet issu applic exhibit drastic behavior base paramet time high lan high san low load drive client adapt server low lan figur perform filter benchmark shown figur execut filter storag server advantag filter run faster cpu resourc show data access filter set determin locat filter run drastic differ grep kernel grep kernel experi data set busi continu increasingli import user applic data filter filter highli select consum larg amount data produc smaller fraction construct synthet filter object return configur percentag input data object highli select filter repres ideal candid execut close long storag resourc vari select cpu consumpt low high defin select filter label low select output data filter high select output input data filter low cpu consumpt minim amount work achiev filter high cpu consumpt simul larg data structur finit state machin text search program time lan san lan client adapt server san figur figur show cach essenti workload exhibit patholog perform insert small record automat enabl disabl client cach place cach object client server result figur show time read filter file number configur set migrat filter client storag ideal filter client migrat filter set achiev perform static locat filter reach optim time requir migrat object bigger fraction total runtim run filter simul load slower storag server make filter run storag server filter execut faster client correctli detect case filter client fourth set move low deem worthi migrat recal migrat threshold note appli estim benefit comput real benefit adapt data access pattern issu cach distribut file system yield dramat reduct storag access latenc avoid slow client total amount memori reduc load server enabl cach yield opposit effect access pattern show appropri migrat cach object data access pattern monitor experi cach provid mobil cach object applic insert small record file store storag server insert requir read block server instal entir block origin block write small record block requir transfer entir content block server applic read cach data desir cach resid client carri experi evalu impact applic access pattern tabl applic insert record file insert write record random locat file tabl applic read record random order block size larg work set applic record experi run warm cach result shown figur locat cach server insert benchmark faster client faster client san optim lan san case differ due rel length cach migrat late san case run multipl observ tabl scan benchmark highlight benefit cach applic workload exhibit leav cach cut execut time cach server lan san test adapt content share data issu filesystem cach namespac distribut improv scalabl content share object client execut object access yield higher scalabl cach local content share object execut effici client invoc serial local avoid overhead retri network kind adapt solv perform cliff share distribut file cach client write rang file happen share common invalid traffic degrad perform server prefer experi chose workload perform directori insert share namespac content benchmark directori present hierarch namespac unix filesystem implement object graph shown figur client access disjoint part directori namespac concurr conflict optimist scheme concurr control check perform isol object section work directori object client maintain cach directori access frequent make directori read fast directori updat minim cheap metadata lock messag perform offload bulk work server result scalabl free storag devic execut demand workload compet client content number retri cach invalid directori object potenti latenc oper content desir directori object migrat storag devic serial client updat elimin retri construct benchmark evalu respond level directori content high content client insert file share directori low content workload client directori directori isol atom storag server client directori network timestamp check cach callback log client client directori isol atom serial server high content directori storag server figur figur show directori updat multipl contend client distribut directori manag client benefici low high content result retri directori oper object move storag multipl client request servic thread serial local cost multipl retri client insert file privat directori result shown figur reduc execut time high content workload migrat directori object server lan optim optim locat directori object host optim high san case worst estim move closer isol object make retri cheaper adapt quickli lan case estim benefit greater observ retri traffic san case decid migrat object low make lan san migrat directori object server migrat start benchmark cold instal read case content applic run estim migrat directori object storag server avoid latenc lan san network time high lan high san low lan client adapt server low san figur figur show time execut directori insert benchmark level directori content migrat directori object fourth case insert scan insert scan tot client adapt server opt tabl phase tabl show perform applic static case applic conclud scan tabl show complet time phase scheme fast model estim instal read network cost limit result show static client storag server configur san case differ migrat threshold benchmark exhibit case placement client popul limit size note directori object client migrat server time server decid migrat base estim migrat benefit client correct ensur object oper verifi occur timestamp order underli object adapt applic phase issu establish optim placement depend system workload note characterist chang time system concern characterist vari algorithm lifetim applic applic rare exhibit behavior consum resourc rate lifetim applic chang phase number point execut input user file result algorithm properti applic make compel case dynam function reloc experi explor revisit file cach run benchmark insert conclud scan phase goal determin benefit estim server applic chang behavior move server recov bad histori quickli achiev adapt applic exhibit multipl contrast phase result tabl show migrat cach locat base behavior applic time migrat cach server insert phase cach object server detect cach client static choic lead bad perform altern phase outperform static compar client compar server case optim row refer minimum execut time pick altern client server case approxim slow optim extrem scenario phase fairli rapidli adapt competit issu share storag server resourc rare dedic serv workload addit complex address provis storag server resourc compet client reduc global applic execut resolv competit object execut quickli server favor object deriv greater benefit experi run filter object file lan filter deriv benefit execut storag server filter produc data filter select output data consum storag memori resourc restrict support filter time result figur show cumul progress filter migrat made select filter start shortli migrat storag server start select filter shortli migrat highli select filter kick back origin node slope curv show filter server run faster filter deriv benefit select filter migrat server notic delai estim benefit close configur threshold longer histori window amort migrat cost longer window result migrat occur sooner histori window long captur iter object statist collect repres applic behavior place overhead compar tradit implement filter implement run run slower implement directli atop unix case migrat occur patholog result perform static part overhead elimin optim implement cumul block process filter move server filter move client filter move server filter filter time figur figur plot cumul number block search filter versu time competit resolv algorithm successfulli select filter filter execut storag server placement case applic complet behavior decid migrat induc migrat back origin chang behavior start cycl applic node incur addit migrat cost linear migrat histori window worst case behavior bound notic object rapidli back locat anchor default placement applic termin demonstr optim function placement depend system workload characterist imposs predict applic design instal time dynam approach function placement continu adapt system base resourc usag avail measur demonstr placement decid base monitor applic system oblivi function implement preliminari evalu show prototyp improv applic time encourag result futur approach richard david shepherd christoph anonym review valuabl feedback refer saltz activ program algorithm evalu eighth intern confer architectur support program languag oper page san oct gold highli concurr share storag intern confer distribut comput republ apr gibson dynam function placement activ storag cluster technic report school comput carnegi mellon june cluster make fast case common sixth workshop parallel distribut page bernstein goodman algorithm concurr control distribut system confer larg oct chamber safeti perform spin oper system symposium oper system page copper dec livni technic summari technic report comput scienc oct dougli ousterhout transpar process design altern implement aug meet virtual machin symposium oper system design implement octob page oct kossmann perform tradeoff queri process acm sigmod intern confer manag volum acm sigmod page june scalabl storag system usenix extrem linux june grai agent flexibl secur mobil agent system fourth annual page juli heidemann develop layer acm transact comput feb ii finkel servic technic report depart comput west scale perform distribut file system acm transact comput feb hunt scott automat distribut partit system symposium oper system design feb kaashoek mobil comput rover toolkit ieee transact special issu mobil mar black mobil emerald system acm transact comput jan hellerstein case intellig disk sigmod record special interest group manag nelson extens file system spring acm symposium oper system page dec flexibl file system base composit acm transact comput aug michel van dam experi distribut process graphic system comput graphic page juli lee safe kernel extens check symposium oper system design faloutso activ storag data mine multimedia vldb confer larg data base august seagat pathwai intellig network press hohl mole java base mobil agent system ecoop workshop mobil object page juli program languag mobil code acm comput sept anderson fault isol acm symposium oper system page dec 
ML/48 6 learn tag inform extract text fabio abstract algorithm adapt inform extract learn templat fill rule insert sgml tag text base cover algorithm learn rule gener instanc tag corpu test scenario inform domain languag experi report excel result respect current state art introduct gener agreement main barrier wide commerci difficulti adapt system applic domain year interest appli machin learn inform extract text interest applic adapt web page inform domain email build fulli autom system due side econom relev applic side applic complex natur languag fulli autom current state art paper present prototyp adapt scenario user corpu tag test inform domain extens web page free text introduc report result task inform domain languag final compar state art system present futur work ongo industri applic present induc symbol rule task base cascad adapt modul perform text entiti recognit templat fill rule formal induct algorithm modul paper report rule induct templat fill work test product mode mode work requir input collect text analyz morpholog disambigu part speech tagger output text augment sgml tag indic presenc extract set templat summar text content oper train mode requir input text mark sgml tag locat inform la ricerca povo itali futur depart comput unit kingdom extract output produc usual set tag rule extract statist accuraci process train corpu test mode requir collect text tag sgml tag produc standard set statist abil locat inform test corpu produc templat rule induct learn insert sgml tag gener exampl found tag corpu learn introduc tag independ close tag make adapt recogn slot filler learn multi slot rule rule induc rule pool contextu rule correct rule induct induc rule pool algorithm start select posit deriv initi rule posit exampl tag found corpu rest corpu consid pool neg exampl initi rule pattern window text surround tag word word text intern repres featur vector token text lemma lexic categori case inform list user defin class specif word dictionari initi rule gener relax constraint reduc pattern length relax constraint term pattern tabl gener test corpu posit neg match record accuraci score wrong match instanc rule report accuraci cover posit cover part error rate threshold gener discard retain rule part rule pool rule enter rule instanc cover remov posit exampl pool rule induct continu select instanc write rule pool posit exampl void find contextu rule effect rule pool extract inform limit matter fact set rule appli rule deriv seed cover portion input constraint present index token lemma cat case sem tag art seminar seminar noun prep digit pm pm verb tabl start instanc seminar pm test corpu high reduc recal mean instanc gener correctli cover number cover instanc left goal step find set rule recal obtain rule lose step reconsid rule select rule discard report low report high recal idea find set rule reliabl restrict situat rule insert miss test context start tag left context tag sequenc encount train corpu effect program david illinoi pm hall low recal rule discard rule insert tag test context identifi tag left context identifi tag contextu tag applic rule restrict gener restrict applic rule specif linguist context envisag case side model specif pattern recurr specif context seminar announc titl frequent deriv rule produc spuriou result appli sequenc mention contextu tag correctli model sequenc note context model sequenc tag captur gener pattern side contextu index token lemma cat case sem tag digit pm tabl gener rule tabl word drop constraint categori capit relax lemma capit relax tag order context tag rule tag word select good rule tag sequenc capit word word rule good rule constrain presenc tag left context mean rule mean organ tag put tag word word correct rule appli rule pool contextu rule test set report high recal locat inform slot filler boundari detect typic mistak part time express aspect common approach adapt report slot reason induc rule shift wrongli posit tag correct posit learn mistak made appli rule pool correct rule train corpu shift rule wrong tag posit distanc correct posit shift rule induc algorithm tag pattern match tag produc previou step rule shift correct shown tabl constraint relax perform usual produc rule test corpu tag previou step posit neg shift correctli assign count shift rule accept report accept error rate threshold token lemma cat case wrong tag correct tag prep digit pm pm tabl correct rule pattern reduc present longer pattern semant field term extract inform test phase inform extract contextu correct valid rule induc step step rule pool initi tag text contextu rule appli context introduc tag contextu tag appli tag tag gener contextu rule match tag insert contextu rule correct rule appli order correct error final insert tag valid subsect valid tag insert algorithm valid mean produc start tag close tag task strategi uncoupl tag produc gener case uncoupl tag consid spuriou remov case contextu inform order insert miss tag system sequenc tag found train corpu order valid extract inform order cover addit miss inform zone highli regular messag post inform typic sequenc field rapier rec prec rec prec id titl salari compani recruit state citi countri languag platform applic area slot tabl result experi job domain half corpu half test add miss tag post messag discrimin spuriou uncoupl tag correct uncoupl tag check text tag uncoupl tag assign statu correct tag coupl default strategi zone time zone mutual text zone reason miss put tag strategi made abil learn separ start end tag end process system produc set contextu rule shift rule result test corpora inform cmu seminar announc corpu austin job announc itc seminar announc develop experi perform train subset corpu hundr depend test learn rule unseen text cmu seminar announc english corpu announc written english corpu develop dayn freitag cmu made rise initi task consist uniqu identifi speaker start end time locat seminar corpu locat tabl show accuraci obtain compar system system test task discuss accuraci calcul muc task recogn speaker respect recogn locat recogn rapier rec prec rec prec speaker locat slot srv whisk rec prec rec prec speaker locat slot tabl result experi cmu seminar domain half corpu half test task concern job announc written provid mari elain califf made rise initi task consist identifi messag job titl applic salari offer compani offer state citi countri job program languag platform applic area requir year desir year requir desir post date result obtain task report tabl obtain system rapier whisk task test collect seminar announc discuss list articl written mix figur task consist identifi titl date time seminar seminar post task identifi singl occurr order highlight result read compar cmu seminar task requir occurr templat slot fill gener similar cmu titl identif complex task matter fact locat repeat corpu finit time author proper easili identifi titl slot complet content determin priori term syntax content titl span line seminar concern comput content restrict titl identif layout shown text titl fact figur whisk whisk obtain lower accuraci rapier newlin speaker featur complex itc seminar announc task bilingu relev limit possibl morpholog po inform languag italian italian morpholog richer languag italian preprocessor mean part written english return wrong lemma italian plural lemma return lemma morpholog find categori return po tagger english part wrong contribut preprocessor case worst case introduc accuraci determin filler task version italian preprocessor mention test accuraci version preprocessor relat research provid excel result knowledg result obtain corpora mention concern symbol approach result compar obtain multistrategi version srv main featur separ recognit tag independ relat start end tag exploit contextu inform opinion higher recal convert strict threshold separ start end tag easier gener rule write rule learn strategi order learn pattern regular express exampl algorithm write rule reason good result respect system stricter contextu inform experi show gener introduc relev error reason good result reli correct step report recognit slot similar experi accuraci obtain correct rule account accuraci speaker cmu experi correct pai job experi due open filler respect slot job domain close compar posit current state art inform domain section compar algorithm test testb seminar job whisk srv rapier system text form induct logic program induc rule corpu tag target templat slot filler consid target field token target phrase token rapier rule match semant morpholog constraint induct figur itc seminar announc start specif rule match target slot train pair randomli chosen rule relax gener gener process repeat progress made obtain result domain system test reason accuraci introduct correct step mention type initi rule similar singl rapier singl slot system concept real differ gener process rapier rule compress strategi gener rule peculiar appendix opinion risk gener make contribut final result harmon produc spuriou result run time rapier compress algorithm complex smart align methodolog order simplifi process hand rapier rule formal power kleen star option oper whisk possibl gener rule extract system regular express extract pattern test structur free text whisk write rule whisk cover algorithm induc rule find po act cor par mi prec rec street citi provinc zip email telephon fax tabl applic train text test text averag length word po correct answer system produc act answer provid system cor correct answer provid system par partial correct answer provid system partial overlap correct wrong answer provid system system select part text mi inform miss system system miss spuriou answer provid system inform present system provid gener rule cover seed ad term special perform level coverag accuraci longer met activ learn reduc tag corpu conjunct parser free text analysi inform domain whisk report lower result respect low impress due strategi gener exampl matter fact problem gener mention system learn singl slot rule worsen rule express exampl mention found context model whisk rule explan low recal obtain whisk cmu seminar task version whisk opinion opposit contextu inform larg step low high whisk reli concept sentenc inform domain seminar concept sentenc unclear gener contextu inform sentenc boundari inform rule worsen problem scarc gener larg inspir whisk initi idea develop bottomup algorithm design found problem mention global contextu inform necessarili relat sentenc spuriou specif valid strategi real differ whisk whisk pattern process sentenc stop absenc stop process add recal domain whisk reach result whisk test type text futur experi comparison srv target extract convert extract problem phrase potenti slot filler system adopt approach combin evid rote naiv classifi relat rule learner rote learner simpli compar phrase list correct slot filler found train naiv classifi comput estim probabl token phrase found correct slot filler relat learner induc manner set constraint length phrase exist word srv complet approach respect compar point view result similar percentag slot intuit result result slot filler result interest understand perform speaker field futur work paper present cover algorithm adapt inform extract algorithm test templat fill inform domain show state art accuraci prototyp base algorithm evalu number applic www wrapper induct joint project main italian internet portal basi search engin extract integr inform web industri project area pharmaceut research test extract inform person canadian compani tabl report preliminari result applic experi mention paper algorithm templat fill separ test algorithm text zone entiti recognit final prototyp includ complet architectur modul expect result mention improv zone ne recognit expect improv accuraci recogn filler involv proper locat ne recognit zone deriv rule rule specif text improv term accuraci domain introduc addit level contextu addit futur work algorithm involv improv rule formal express order includ form option oper gener rule explor form activ learn order reduc burden text tag tag task requir half person dai graphic interfac algorithm result improv addit activ learn untag corpu rapier whisk includ declar strategi process templat merg task straightforward scenario inform case applic free plan integr modul order overcom limit refer descript entiti system mari elain relat learn techniqu natur languag inform phd univ texa method inform ai fabio home page fabio alberto inform extract european confer io press aaron messag understand confer score softwar user extract applic gener learn proc fifteenth confer artifici intellig dayn machin learn inform extract inform phd carnegi mellon sra system proc sixth messag understand induct inform proc intern artifici annot descript sift system induct inform proc confer autonom learn discoveri ellen construct dictionari inform extract eleventh nation confer artifici pp repositori onlin inform sourc inform extract inform scienc institut inform extract rule semistructur free machin 
ML/1981 6 valu similar distanc box mathemat dept physic comput faculti aristotl thessaloniki gr greec juli abstract start point paper introduct measur fuzzi set fuzzi set measur valu interv measur valu boolean lattic view fuzzi valu relat fuzzi set relat fuzzi order addit number properti author postul axiomat measur defin valu measur similar fuzzi set valu distanc function fuzzi possess properti analog similar distanc function fuzzi transit introduct measur call relat fuzzi set subset measur valu case correspond crisp case fuzzi relat concept measur similar fuzzi measur similar relat fuzzif crisp relat paper introduc denot denot fuzzi set includ fuzzi set introduc relat similar measur denot distanc fuzzi set main differ work author similar distanc valu total order subset real interv rang boolean lattic partial order similar distanc valu relat fuzzi set sake breviti refer quantiti valu total order set quantiti valu partial order set terminolog obviou sequel measur symmetr transit order relat number attract properti author postul axiomat measur similar remark hold similar distanc briefli review relat previou work semin paper definit fuzzi set crisp relat fuzzi set includ includ fuzzi set popular concept fuzzi set essenti fuzzi analog condit probabl paper measur section work instanc interest variat measur paper measur section author axiomat approach studi provid list properti measur satisfi examin measur famili measur possess properti prime approach sinha list properti measur proce introduc measur properti point view author obtain measur fuzzi implic oper relat approach transit measur studi issu transit studi measur introduc axiomat connect implic oper combin number paper young combin approach connect measur fuzzi entropi young exampl measur list literatur examin erti satisfi measur connect work fuzzi implic oper young present applic measur tune fuzzi logic rule al introduc famili fuzzi cation oper form gener show measur obtain famili satisfi sinha axiom measur fan al discuss connect fuzzi entropi fuzzi implic comment axiom list axiom give condit check function measur present applic measur cluster valid measur discuss point valu total order set introduc measur valu partial order set interv valu fuzzi set relat implic oper view measur author classifi measur dissimilar gener scheme gener measur elementari properti gener author local measur similar global mention connect similar measur discuss fuzzi similar receiv attent fuzzi literatur discuss small number point view similar collabor issu seri paper axiomat view similar measur fan axiom distanc similar liu altern set axiom wang al critic work present modifi definit similar explor connect work wang adopt similar axiom introduc similar measur discuss relat valu lattic mention similar relat context return topic mention view author introduc fuzzi measur crisp set crisp set develop extens methodolog measur cluster classif applic point practic work fuzzi similar measur valu total order interv notabl except interv valu similar gener discuss valu relat found preliminari fuzzi crisp set subsequ discuss make fundament set denot special structur properti assum countabl assum order element total order set denot paper result remain valid case finit set element denot letter element denot equal symbol usual minimum denot maximum denot introduc negat element defin usual subtract real number de morgan bound distribut respect respect oper satisfi de law complet lattic total order concern fuzzi fuzzi set identifi membership valu fuzzi set function fuzzi set denot empti fuzzi set denot fuzzi set denot case repres fuzzi set vector notat element belong set denot au famili fuzzi set fuzzi powerset denot simpli exampl fuzzi set crisp special case fuzzi set identifi membership valu fuzzi set function θ underscor fact set crisp denot greek letter famili crisp set crisp powerset denot simpli exampl crisp set θ φ lattic remark total order de morgan lattic defin order element obtain order denot danger symbol defin au easi order time direct product lattic time partial order exist element denot product complet complet inf oper element respect obtain min max oper denot danger defin au bu notat function denot denot danger defin au bu complement defin term continu set previou exampl au write similarli φ easi prove de morgan lattic easi prove boolean lattic boolean complet lattic fuzzi relat sequel fuzzi relat subset fuzzi relat interest function form adapt definit context definit fuzzi relat call reflex iff definit fuzzi relat call symmetr iff definit fuzzi relat call iff definit fuzzi relat call iff definit fuzzi relat call iff inf remark definit inf exist complet lattic definit fuzzi relat call fuzzi similar relat fuzzi symmetr definit fuzzi relat call fuzzi order relat note function domain rang denot point notat measur definit properti definit measur fuzzi denot defin iff au bu continu set previou note motiv definit set set defin term valu true conclud includ maximum differ definit altern definit measur set aggreg singl real rang tradit measur total work collect quantiti valu slightli unusu place argument function notat prove advantag sequel partial order set valu partial total exist set includ includ compar part mention refer quantiti valu total order set quantiti valu partial order set case finit cardin boolean vector size case infinit consid vector vector space term intend invok connot addit partial order oper defin section introduc vector term proposit describ attract properti proposit fuzzi order fuzzi relat proof reflex au case au bu bu au au bu au bu bu au lead contradict case au cu cu impli au bu proce show number addit author axiomat postul desir measur proposit au proof prove properti frequent make fact rang bu total order set au bu au bu gener true partial order similar fact kei import argument present sequel assum au similar manner prove assum au au au au case au bu au bu au bu au assum bu cu case au au au cu assum bu cu case cu cu bu au au bu cu du au cu bu du au cu bu du au bu cu du au cu bu du au cu bu du immedi obtain substitut obtain yield desir result substitut obtain yield desir result substitut obtain yield desir result substitut obtain yield desir result substitut yield remark properti express monoton properti relat combin monoton worth remark measur enjoi properti place properti list section proposit θ θ θ proof immedi comparison measur section briefli compar measur introduc author list author reason measur compar properti introduc paper axiomat measur sinha reproduc present axiom tabl remark order note adapt notat author mention parallel notat symbol denot scalar measur vector properti tabl refer measur total order maximum element minimum element remark author mention present altern list tabl meant give repres view consid properti measur note term properti list tabl independ impli impli stronger form stronger form alter order present properti column tabl list id number final column list properti proposit column list actual axiom properti remain column paper properti properti crisp set au bu tabl measur axiom paper tabl properti tabl enjoi analog properti tabl exhaust list properti reason offer pretti complet coverag omit crucial sinha remark choic relev reason measur larg reason properti addit enjoi case scalar measur literatur list specif exampl measur list popular notat au omit sinha properti invari measur respect shift similar properti easili prove respect relabel set element au α bu measur measur introduc wide refer remain measur measur list measur introduc transit fuzzi measur involv drastic modifi definit transit requir easili satisfi measur valu total order rang hand satisfi addit larg number measur similar distanc defin term proce defin similar distanc term similar definit measur similar denot defin domain partial order set result possibl incompar similar measur continu set previou definit set includ maximum vice set ident maximum similar basic properti proposit proposit au bu proof bu bu au properti present proposit proposit fuzzi similar relat symmetr transit proof reflex proposit symmetri set θ ψ φ γ ω proposit θ ψ proposit au au quantiti proof similarli prove au au assum au easi au similarli au au case au cu conjunct au bu impli au bu prove similarli transit proposit similarli prove obtain substitut proposit impli prove similarli note case au au bu au bu au au bu au bu au bu au bu interchang role obtain prove show equal symmetri transit substitut obtain prove interchang role immedi prove proof complet prove similarli previou step step complet proof remark properti relat concept properti relat modular concept relat present remark section proposit θ θ θ θ θ θ θ proof θ θ θ case au cu bu cu au cu bu cu distribut au bu au cu bu cu au cu bu cu au bu remark proposit relat concept introduc distanc definit distanc denot defin continu set previou proposit au bu proof au definit distanc defin complement similar complement set larg small distanc distanc perceiv total nonneg definit unusu fact basic characterist distanc outlin proposit proposit proof definit proposit remark partial order analog triangl inequ addit triangl idea boolean valu distanc instanc book idea appli boolean lattic call space domain metric distanc function space metric case space proposit interest proposit φ φ proof φ case case φ proof present properti proposit au au quantiti proof definit proposit remark properti relat concept convex abstract metric space scalar distanc function point straight line point iff direct analog case euclidean complet space fuzzi iff shown satisfi condit properti special case condit convex defin collect fuzzi set convex iff point remark properti relat lattic modular modular lattic posit valuat defin distanc detail easi prove analog equal proposit θ θ θ θ θ θ θ proof definit proposit introduc valu measur establish number properti properti analog postul fuzzi measur find attract turn fuzzi order defin similar distanc fuzzi set conclud discuss futur research direct order lattic order relationship fact order relationship denot suggest manner write θ θ write easi show fix crisp preorder problem natur problem defin lattic manner famili cut develop lattic theori implic relationship set logic implic discuss part measur view valu implic oper connect report parallel work author obtain measur fuzzi implic style implic oper satisfi properti analog axiom implic oper connect condit probabl worth investig scalar measur wai analog measur section interest idea present paper defin probabl measur aggreg interest investig exist measur measur lie extrem end spectrum scalar sion measur aggreg inform element set singl preserv inform extrem aggreg extrem local prove rout arriv measur defin partit defin vector measur scalar measur depend valu uk detail aggreg singl local inform preserv mention approach similar measur author view similar measur relat result similar membership local relat relat approach implic oper discuss transit paper local prefer aggreg find interest explicit recognit similar determin term local relat approach perform aggreg step refer lattic american mathemat colloquium vol power set fuzzi implic fuzzi set lm theori applic distanc geometri clarendon fuzzi concept fuzzi set gener measur comparison fuzzi set vol pp grade fuzzi implic fuzzi set pp grade fuzzi set applic approxim reason base fuzzi int approxim ba ha introduct lattic order cambridg xie pei fuzzi set fan note similar measur proxim fuzzi set jc fodor aggreg fuzzi binari fuzzi set fodor applic aggreg procedur definit measur similar fuzzi fuzzi set vg framework fuzzi learn techniqu comput intellig ed lc jain boca crc vg lattic neurocomput connectionist scheme versatil learn intern journal comput vol pp gj klir yuan fuzzi set fuzzi prentic neural network fuzzi system kundu theorem fuzzi fuzzi set xc distanc measur similar measur fuzzi set fuzzi set pp ht ea walker fuzzi logic crc press fuzzi partit fuzzi fuzzi set cp approxim fuzzi system fuzzi set cp ni compar assess measur similar fuzzi fuzzi set cp ni similar measur fuzzi set fuzzi relat fuzzi set vg lattic neural network hybrid model ieee transact neural vol order relat valu fuzzi relat fuzzi set sinha er dougherti set theori fuzzi set vol pp lattic valu fuzzi fuzzi set system wang similar measur fuzzi set fuzzi set compar studi similar fuzzi set transit fuzzi power set math anal vol implic oper theori fuzzi power fuzzi set young fuzzi set la inform 
IR/1109 5 steve lawrenc context web ieee data engin volum number pp context web search steve lawrenc nec research independ nj abstract web search engin gener treat search request isol result queri independ context user made request search engin make context explicit implicit context inform implement addit function restrict context greater context web search competit web introduct web increasingli repres area societi inform web author organ million expect contrast tradit inform retriev web term content structur current web search engin similar oper tradit inform retriev system creat index word return rank list document user queri web search engin good return long list relev document user method improv rank search result result return search engin valuabl user document valuabl depend context queri previou experi inform current request user compani sell technic detail user site search engin googl fast make inform easili access wide web gvu studi show peopl search engin locat inform search engin consist rank top site access web major web search engin limit index fraction publicli index index document authent requir document search index site equal popul task perform search servic increasingli import understand context search request web search engin gener treat search request isol result queri independ context user made request context inform provid user form keyword ad user homepag individu add keyword queri provid context form difficult limit add context inform search request search engin specif request inform ad explicit context inform inquiru project nec research institut request context form categori inform desir addit provid keyword user categori introductori inquiru metasearch engin oper layer regular search engin inquiru queri context attempt context inform find relev document regular web search engin context inform select search engin send queri modifi select order polici queri research paper send multipl queri search engin queri transform addit keyword improv find research paper queri ident origin case transform success inquiru proven highli effect improv search result categori recent research relat inquiru includ learn method automat learn queri modif automat infer context inform inquiru greatli improv search requir user explicitli enter context inform search context automat goal watson project watson attempt model context user inform base content document edit microsoft view internet explor document user edit analyz heurist term weight aim identifi word indic content document inform font size weight word user enter explicit watson modifi queri base content document edit forward modifi queri web search automat ad context inform web search addit explicit watson oper continu document web relat document user edit view mode oper similar remembr agent remembr agent index file email messag research continu search relat document user edit document emac editor relat project margin note rewrit web page includ link relat person haystack project aim creat commun interact person inform autonomi program automat suggest content web local base document user read edit relat agent learn user interest profil recommend web page fab letizia webwatch syskil webert person search step complet person search search engin previou request inform tailor result request rank link professor comput scienc statist highli link famou individu interest person search engin server search engin googl track user previou queri select inform infer user interest user search comput scienc relat materi homepag comput scientist rank highli queri user search person search servic track document edit view order obtain model user interest servic local access larg scale index limit function servic rank homepag comput scientist highli queri search servic return page maximum number result client retriev client modifi queri retriev document relat difficult entir interest user watson exampl person web search engin watson extract context inform current document user edit view cost run larg scale search engin person major web search engin major search engin light provid alert servic notifi user page match specif queri advanc comput resourc make larg scale person search feasibl time internet compani devot substanti amount storag individu user compani offer free disk storag user import problem person search servic user expect consist receiv result person search engin return result user engin learn user import address privaci user limit storag person inform search engin compani guess user increasingli common techniqu web guess context user queri search engin excit lyco googl yahoo provid special function kind queri queri excit lyco match artist compani produc addit result link directli artist compani inform yahoo recent ad similar special result type queri stock symbol provid stock quot link compani sport team link team leagu inform exampl yahoo includ car major drug dog tv nation park googl identifi queri street direct link map googl track recent link match articl effect guess user articl explicitli requir user enter context inform stock techniqu guess context relev user easili identifi context interest techniqu limit case potenti context identifi base keyword queri improv guess search context person search engin queri return link list prof michael jordan public scientif user interest comput guess user list public prof jordan cluster search perform northern light relat northern light dynam cluster search result categori user narrow result categori restrict context search engin add context web search restrict context search creat special search engin specif domain thousand special search engin exist servic provid similar function regular web search inform publicli index web fraction index regular search inform regular search engin york time search number special search engin provid function provid regular web search specif domain inform extract domainspecif process researchindex special search engin scientif literatur researchindex free public servic researchindex world largest free fulltext index scientif index articl million citat incorpor featur specif scientif literatur researchindex autom creation citat indic scientif easi access context citat special function extract inform commonli found research articl special search engin inform extract process includ deadlin confer workshop inform newsgroup mail job inform index homepag comput inform extract analysi link sourc order determin geograph locat scope web resourc northern light servic call northern light attempt extract address web incorpor concept geograph scope resourc york time locat york interest larger geograph local york newspap interest search engin automat extract inform web page method inform identifi commun web search engin index inform publicli index web method locat subset web domain flake al recent shown link structur web selforgan commun highli relat page effici identifi base pure connect web commun defin collect page member link insid commun commun definit gener identifi commun size vari level discoveri import central author process govern format link web discoveri identif commun web independ specif word algorithm effici identif commun found method locat commun relat page web locat special search engin thousand special search user locat interest user effort locat special search engin queri serv special servic major web search engin overhead locat special engin great exist method locat special search engin research area method select search engin base user queri gloss maintain word statist order estim queri relat research includ great benefit major web search engin attempt direct user special search engin search engin incent provid servic prefer maxim servic provid size fit limit competit typic search engin view size fit user receiv queri argu model optim serv larger state benefit web equal access inform equal web distribut traffic link site extrem skew approxim power law share traffic link small number popular site evid trend behavior market share popular servic largest convent book retail market largest onlin book retail market share search engin contribut statist prior consum locat store store list phone book number consum locat store search engin imagin web search keyword result site rank popular measur incorpor select rank criteria small percentag peopl search engin find peopl creat link web enhanc bia locat store experi locat item web common experi compar previou mean locat item phone walk note trend concern trend common experi watch number cabl fewer peopl watch program access inform limit search servic queri search store major web search engin return user find list onlin yellow page phone directori search presidenti candid februari differ major web search engin probabl offici candid homepag return page result similar search special polit search engin provid result exist servic prevent bia inform access peopl major web search engin search directori site yahoo open directori unequ delai list site list directori extent effect depend peopl search engin locat kind search engin search servic incorpor incorpor context exist search mitig neg effect access inform web summari search engin make amount inform quickli easili access contribut web societi enorm size fit model web search limit function context web search web search import function search servic increasingli import refer stein inform environ confer inform knowledg gravano extract relat larg collect acm intern confer digit mitchel learn world wide web marko balabanov adapt web page recommend servic intern confer autonom page acm barabasi albert emerg scale random network carol barri identif user criteria relev document topic approach inform retriev phd bharat mr improv algorithm topic distil hyperlink environ sigir confer research develop inform joachim machin learn architectur optim web search engin aaai workshop inform brin extract pattern relat world wide web webdb workshop edbt brin page anatomi hypertextu web search engin seventh intern world wide web kj hammond user interact everydai applic context inform access intern confer intellig user acm press kj anticip inform everydai applic interfac internet inform server world confer internet aac press jai kristian larri marko similar workshop artifici intellig web search aaai garc shivakumar exploit locat inform web page acm web rajagopalan automat resourc list compil analyz hyperlink structur text intern world wide web soumen martin van den byron dom approach web resourc discoveri world wide web junghoo hector garc lawrenc page effici crawl url order seventh web fran eric steve lee gile featur select web applic roc inflect symposium applic san januari hawk server select world wide web acm confer digit page slatteri learn extract symbol knowledg world wide web fifteenth nation confer artifici aaai page wu appli link analysi web search eighth intern world wide web page fran steve lee marco gori crawl context graph intern confer larg vldb septemb lui narayanan shivakumar comput geograph scope web resourc intern confer larg vldb septemb experi select search engin acm transact inform gari steve lee gile effici identif web commun sixth acm sigkdd intern confer knowledg discoveri data page august susan mario gomez intellig fusion distribut search engin journal comput raghavan infer web commun link topolog acm confer hypertext eric gari steve william lee david pennock improv categori specif web search learn queri modif symposium applic san januari eric steve william lee gile architectur metasearch engin support user inform eighth intern confer inform knowledg cikm page kansa novemb eric steve michael william lee gile web search commun accept public usabl center gvu tenth www user survei octob garc paepck stanford internet acm sigmod intern confer manag page garc discoveri internet acm transact lui gravano hector garc gener gloss broker hierarchi intern confer larg page ba je rm strong regular world wide web surf kleinberg authorit sourc hyperlink environ symposium discret page san januari lee fran eric gari steve cristian build nich search engin ninth intern confer inform knowledg cikm novemb kushmerick wrapper effici express workshop ai inform doorenbo wrapper induct inform extract ijcai page steve kurt lee gile index retriev scientif literatur eighth intern confer inform knowledg cikm page kansa novemb steve lawrenc lee gile context page analysi improv web search ieee internet steve lawrenc lee gile search world wide web steve lawrenc lee gile access inform web steve lawrenc lee gile search gener scientif inform access ieee steve lee kurt bollack digit librari autonom citat index ieee hammond select sourc retriev workshop intellig inform menlo aaai press lieberman agent assist web intern joint confer media media announc top digit media properti site januari rish estim search engin intern confer data stefano histori journal american societi inform billsu syskil identifi interest web site nation confer artifici ivan png competit convent preliminari studi nec research renni mccallum reinforc learn spider web effici sixteenth intern confer machin learn bradlei rhode inform retriev phd institut bradlei rhode margin build contextu awar associ memori intern confer intellig user iui bradlei rhode thad starner remembr continu run autom inform retriev system intern confer practic applic intellig agent multi agent page salton automat text analysi retriev inform comput andrew rosenfeld learn hidden markov model structur inform extract aaai workshop machin learn inform soderland learn inform extract rule free text machin sullivan find search engin februari xu callan effect retriev distribut collect intern acm sigir confer research develop inform page zobel collect select lexicon inspect australian document comput page 
IR/1376 5 decemb automat text detect track digit video li david doermann automat text detect track digit video david doermann languag media process laboratori institut advanc comput studi maryland colleg md mathemat comput scienc nation institut standard technolog md abstract text scene graphic ad video provid import supplement sourc index inform clue decod video structur classif paper present algorithm detect track text compon digit video frame system implement featur extractor feed artifici neural processor extract textual region track movement time extract region input optic charact recognit system produc index keyword text text video digit neural wavelet support effort depart contract mda gratefulli acknowledg introduct avail onlin digit imageri video interest problem index multimedia inform sourc automat manipul effici imag video sequenc manual annot small number keyword descriptor visual inspect human review process accept archiv inhibit abil perform filter retriev problem index understood address tremend success automat document optic charact recognit technolog transcript speech voic recognit technolog typic ascii text index tradit inform retriev techniqu deal visual linguist content video sourc movi sport event content form text text imag sourc complement abil provid essenti supplement inform index retriev address problem detect track text digit video sound close caption provid index inform spoken basic annot inform imag text sport product scene speaker movi program introduct special announc supplement summar visual content type annot render high contrast respect readabl keyword facilit index specif search actor refer stori easili realiz exist access textual content standard video code extend scope video stream multimedia effort motion pictur expert group part joint intern organ standard intern engin produc standard standard video encod static background move foreground ultim annot textual content object make easier extract index tool standard wide implement expect accept mainstream year extract textual content render part video text scene extrem valuabl index background embark gener subject text extract specif aspect problem discuss synthet natur sourc textual content video frame review relat work present approach scene text graphic text high text divid scene text graphic text scene text scene captur record devic integr part imag consid sampl real world exampl scene text includ street text truck write shirt text typic scene applic surveil read text gener index retriev except constrain text symbol identifi player vehicl scene text difficult detect extract virtual unlimit number shape color graphic text mechan ad video frame supplement visual audio structur close relat subject scene text exampl graphic text includ keyword time locat peopl sport score descriptor typic simpl produc intent read viewer graphic text number function differ domain text reinforc vital inform product claim provid disclaim sport text identifi specif provid game relai statist graphic text identifi kei featur locat speaker provid synopsi topic provid visual summari statist inform movi text product act case caption languag translat research area identif graphic text relat work relat domain work extract text road sign plate librari book www imag scene imag isol video frame method broadli classifi type connect compon base unlik binari document scene imag video frame multivalu multivalu imag decomposit requir connect compon extract gener idea method multivalu imag pixel valu integ greater set elementari imag fi pixel assumpt text repres uniform color connect compon extract elementari imag heurist restrict compon number align compon line orient identifi text line color color cluster techniqu requir binar process typic depend complex color cluster binar method method locat text difficulti text complex background touch text graphic object approach textur analysi method gabor filter gaussian filter spatial varianc locat text region jain describ method separ text imag area base group gabor filter wu present text extract system base gaussian filter treat text distinct textur pixel origin imag repres featur vector consist energi calcul filter imag cluster cluster featur vector text detect digit author analysi add miss charact delet incorrectli identifi region shim detect text consecut frame examin similar text region term intens shape extract text frame check region correspond text frame simpl region match frame filter incorrectli identifi text region work address problem track text find tempor correspond detect text digit video text video frame save recognit literatur object track extens includ face track human bodi track vehicl track medic imag agricultur autom detail survei track algorithm motion found literatur track text approach text detect track digit video view detect track problem multipl text block time move direct field view unlik typic track problem static background move combin static move text background treat background detect track text simpli motion analysi special case movi make motion inform segment motion inform suffici text detect scheme requir conjunct track simplist solut text track perform text detect frame frame match text block consecut frame close frame text detect phase absenc context prohibit make fact text remain scene number consecut frame reduc complex perform text detect process period track process text detect scheme base observ text region typic textur properti surround area textur similar frequenc orient make wavelet reason candid hybrid network segment segment text region facilit detect text pyramid imag gener origin imag halv resolut level extract text region level track process detect process figur scheme text detect track digit video pyramid input imag wavelet decomposit featur extract statist normal neural network layer bp scan window pyramid output imag integr imag scale figur architectur detect text video frame extrapol origin scale text ssd squar base track modul start track detect text make text contour stabil track text track robustli remaind paper organ describ text detect scheme section text track scheme section result present section section discuss applic work figur video binar manual pick threshold text detect video frame compar text detect text detect digit video challeng text digit video complex make difficult connect compon method figur show typic video frame figur binar version good threshold binar imag show charact compon background text digit video render low resolut document scan dot inch result charact occupi area larg pixel video frame limit spatial resolut low charact size approxim pixel effect low resolut text color cluster result scatter text text compon tend touch make detect difficult video natur scene leav tree grass field textur similar text featur text overlap illustr collect text block block linear discrimin analysi project subspac scatter maxim scatter benefit classif project result show text overlap featur space learn classifi text effect cluster techniqu artifici neural network natur figur distribut text subspac lda project text overlap choic classifi abil learn neural network approxim nonlinear function train success neural network relat problem motiv reli neural network classifi identifi text region methodolog small window scan imag classifi window text neural network address issu relat ffl decomposit ffl featur extract ffl classif neural ffl text identif scale space decomposit analysi scale space method identifi spatial frequenc content local region imag wavelet imag provid success approxim imag downsampl abil detect edg figur wavelet decomposit video origin imag filter filter creat success approxim imag detail signal term textual content easili imag decomposit shown figur origin imag wavelet decomposit note text region show high activ result local wavelet locat edg yield larg wavelet make text region detect high frequenc high frequenc compon import recognit jone palmer shown wavelet close repres profil neuron striat cortex depend choic suitabl wavelet decomposit level base criteria wavelet chosen adequ local detect line text region addit imag decomposit implement simpl mask weight sum make wavelet effici improv observ type base experi scale wavelet function wavelet written figur mask lowest vertic high horizont high high frequenc horizont vertic direct wh sequenc equat valu valu equat imag repres algorithm obtain wavelet transform ll lh hl hh imag decomposit process convolut imag kernel figur comput effici obviou featur extract select task featur extract select solv close form method featur select explain section featur extract extract featur wavelet decomposit imag shown figur text textur background detect featur central moment subblock calcul central moment subblock written featur comput imag origin window size maximum decomposit level pixel left imag fourth level featur salienc featur salienc defin measur featur abil impact classif benchmark compar featur salienc probabl error denot comput probabl error singl featur salienc metric featur measur mlp sampl rang input valu empir result metric similar rank error rate measur salienc error rate comput determin featur yield adequ separ class practic error rate estim finit set label sampl class typic estim posterior probabl class assign sampl class map decomposit level imag ll hl lh hh ll hl lh hh ll hl lh hh tabl error measur featur salienc decomposit scale probabl estim percentag sampl misclassifi appli map rule posterior estim estim error rate interfac implement collect text sampl sampl interfac user load imag simpl oper label text region region collect sampl collect text block block sampl class train remain sampl test tabl list featur drawn tabl featur larger valu featur level explan imag size level inform valu gener larger moment order imag smaller imag decomposit level impli moment calcul imag decomposit level order verifi test data wavelet ignor level level provid addit discrimin tabl result wavelet tabl result wavelet exist small draw level subimag expect varianc order moment decomposit ll hl lh hh ll hl lh hh tabl error wavelet level subimag expect varianc order moment decomposit ll hl lh hh ll hl lh hh tabl error wavelet wavelet case compar result tabl wavelet outperform daubechi wavelet problem featur select criterion featur featur set small featur maxim classif result obviou reason reduc featur set effici train smaller number paramet author address impact requir train sampl input featur curv show number train sampl requir grow exponenti number featur salienc analysi section featur set rank twelv featur order error shown tabl featur rank tabl rank featur order denot moment repres decomposit level repres scheme select featur featur tabl train classifi calcul classif error add featur featur repeat process step featur ad twelv featur limit method inabl deal data neural network classifi analysi salienc chang number input paramet neural network number input chang group adjust number minim classif error reach figur illustr relat classif error number featur classif result achiev twelv featur drop featur balanc accuraci process time conduct window video frame account princip process time text featur tabl select classif configur neural network consist layer input hidden node output node train neural network select neural network sampl collect featur analysi cover featur space extens easi repres sampl difficult repres sampl span vast space deal bootstrap method recommend sung poggio idea train sampl collect train train classif error number featur figur test set classif error number featur creat initi set train sampl includ complet set text sampl portion sampl train network sampl run system video frame text add imag block network incorrectli classifi text sampl set repeat step accuraci converg neural network map block real text figur show distribut neural network output test text block train threshold reason choic window text text class separ neural network classifi text identif train neural window scan video frame classifi window text larger step window smaller number window process refin result video frame size system classifi window window move pixel sun ultra step size window dont text line split window output neural network figur output distribut train neural network test text block solid line repres text dash line repres figur origin classifi label segment text area segment text area bound box gener detect move window pixel time window classifi pixel window label text pixel cover text window label result classif label map origin imag figur video frame figur classifi label map figur figur show extract text region text label small isol area incorrectli label text size consist block filter area bound box text area gener connect compon analysi text window figur result filter area gener bound box result integr depict figur pyramid approach detect text wide rang font size detect step appli bound box map back origin input imag bound box level overlap box simpli merg origin imag text string part text string multipl level box detect level overlap case merg form bound box figur integr result figur figur text region detect integr result postprocess result text detect algorithm set text multipl text line text spatial compact text detect result form block benefit imag resolut enhanc ocr text line provid context block typic good candid track size text line text block move text remain static text move divid text block compact text typic line word project profil extract text element text block avoid difficulti normal text project profil canni edg map imag imag figur set gener text element detect result figur block horizont orient track text track problem overview text track process start text motion digit video simpl rigid motion scroll movi complex motion scene text zoom rotat camera simpl imag match techniqu track text robust techniqu requir goal design scheme figur result text canni edg map detect text horizont project profil gener text line effici track text simpl complex motion arbitrari motion encompass wide rang motion transform gener paradigm estim motion paramet extrem develop literatur research approach deal motion restrict object behavior reliabl visual track object complex environ requir integr visual criterion emploi assumpt incom imag modul select complement modul aid treat text line close set bound box text line correspond boundari boundari interior provid complementari determin shape determin content integr track modul base imag content shape make track process power stabl gener ssd squar modul measur similar imag content interior make inform text contour stabil posit text block boundari rest section give detail implement ssd squar base imag match camera text pattern imag intens complex wai describ imag subsequ imag time obtain move point imag time amount ffi repres affin ffi deform matrix displac vector minim ssd squar match basic track algorithm refer block imag search region imag determin paramet deform matrix displac vector minim ffl solut equat complex qualiti estim depend factor size featur imag amount camera motion assumpt small simplifi equat set deform matrix ident matrix equat simplifi ffl model equat pure translat model search space rang predict locat simpl predict techniqu locat accur text posit current past essenc linear predictor robust model adapt linear predict kalman add comput complex frame number frame frame frame frame number ssd figur model track movi credit initi posit manual track graph ssd consecut frame ssd higher frame text line move complex background pure translat model track case simpl motion static text figur show track movi credit text line track move clean background complex background middl text line correctli track figur show graph ssd pure translat model adequ handl rotat distort long sequenc error accumul point tracker lose target figur show track result track text line confer video sequenc camera scene chang text move complex includ ssd small consecut frame tracker gradual lose target frame text stabil text undergo complex pure translat model track exist elabor track model includ parametr model articul frame number frame frame frame ssd frame number figur method fail distort confer exist scale chang rotat deform linear imag subspac result algorithm reli nonlinear optim techniqu requir minut frame comput text contour stabil track process stabil process implement effici match text posit gener slightli larger text block ffi real text posit includ gener edg map calcul canni edg edg map threshold imag avoid difficulti identifi normal text text appli horizont process edg map group form text block extract connect compon posit repres refin text posit figur show initi match posit figur show refin posit figur text posit refin base contour stabil initi match text enlarg text edg refin posit scheme work text move clean background text move complex problem occur complex text touch graphic object contour extract process oper slightli enlarg box leav extra space contour text contour larger text rigid motion text move complex ssd consecut frame larger pixel valu background chang contour harder extract contour stabil achiev satisfactori result case stop stabil process depend ssd model text move complex background smaller stabil start match reduc complex modul comput cost track larg text line perform match fine hierarch fashion gaussian imag pyramid frame size gaussian pyramid form combin gaussian imag frame frame number ng repres level pyramid size frame level figur multipl resolut base imag match text block pyramid frame imag pyramid form frame match process conduct imag scale match conduct start resolut level contribut determin posit match level search minimum ssd start level window size match point found level search minimum ssd conduct pixel window size process continu finest resolut reach maximum displac support search smaller requir displac doubl level total displac reach finest level figur illustr process level pyramid depend size text block text line small pyramid form match conduct directli origin imag scale text detect region text cross screen figur make tempor tendenc tempor tendenc goal system track text gener tempor tendenc text motion facilit track process text line scroll deduc text line bottom video frame restrict costli text detect process small region text cross frame analysi tempor tendenc track text correctli newscast announc cross bottom tv screen track text line case virtual text line bigger frame separ text line word track word find word text line move horizont direct word continu draw text monitor small area text word implement experi implement tool base method text detect tool text track tool text detect track tool program written run sun workstat oper system solari detect text region singl video frame track text region initi posit combin modul conduct epoch network error epoch figur relat train time detect track sequenc frame evalu perform tool experi text detect collect video frame split train set test set train sampl block collect video frame train set bootstrap method train time depend requir error figur show smaller longer train neural network detect procedur requir sun workstat ultra process frame unoptim code classif featur postprocess imag input output text collect includ scene text graphic text multipl font size evalu text detect system block level segment text block word charact text block text line close shown tabl total text block frame correctli detect algorithm miss error occur primarili low resolut text block frame number text line detect miss detect tabl text detect result figur part identif result text larg font text low resolut scene text text font text font style bottom small text block size text string block misclassifi text block figur exampl attempt ocr overcom problem slight detect segment task direct applic document imag segment chang number node output layer correspond class background figur exampl text block bottom frame detect due low text string detect due small block top frame block middl frame block misclassifi figur slight segment document imag document imag scan magazin segment result document imag washington segment result featur extract scheme number hidden layer node adjust maxim classif result chose train set document washington test system rest document scan magazin cover figur show work layout document complex text track collect ten video sequenc experi text track wide varieti video includ movi tv footbal confer descript video includ static simpl complex tabl conduct experi step studi perform text track manual initi posit text filter effect text detect quantit evalu track accuraci easi lack ground truth data output track result form mpeg video view websit track result shown figur experi show tracker work text simpl motion motion complex background clean video video frame text motion id type number line descript movi credit complex movi credit scroll static sport complex sport complex confer complex confer complex scene cross tv program zoom commerci cross tabl video sourc experi text move arbitrarili complex tracker frame adjust posit text move clean background figur initi posit touch object edg text size track posit deviat frame adjust text grow larg separ map track number athlet jersei footbal game task complic athlet camera motion tracker lose part target frame adjust correct posit averag track time text block frame text line larg match reduc track text text block tempor correspond block perform detect frame extra time requir find correspond block consecut frame applic text track tool ground truth video video process evalu resourc project lab function mechan creat ground truth data perform quantit evalu account cost ground truth video manual track reduc cost significantli user initi box track rest frame user check adjust text figur demonstr perform video sourc initi posit text block manual track movi text zoom displac zoom scene text text scroll vari background track text footbal athlet jersei track text confer room scene track result found mpeg video frame frame frame frame frame frame frame frame figur text detect track movi star war figur languag media process frame frame frame frame part experi combin text detect track modul input sequenc imag output sequenc detect track text region repres mpeg video ascii file record inform text block figur show track result movi star war frame includ scroll text figur show track result text line detect horizont scroll tempor tendenc analysi section divid line word track result view websit process time number text line frame track movi credit time video type text line frame movi credit movi star track frame averag number text line frame track text footbal game text line move detect text tempor correspond text block limit system text track start text detect text detect modul system miss text tracker imag match approxim posit text contour refin posit system track text speed predict predict posit text limit tracker difficulti text move move complex background sport video track athlet text occlud quickli athlet jump rotat applic discuss section discuss applic involv detect track text video index ocr digit video video ocr requir convert text imag format plain text block ocr mention text digit video low resolut limit commerci ocr softwar figur show text block extract video frame manual pick threshold binar imag output ocr softwar text readabl text digit video limit spatial resolut font size approxim pixel typic document imag commerci ocr softwar work scan dot inch translat charact occupi area larg pixel motiv perform text resolut enhanc achiev reason ocr result commerci ocr softwar index retriev improv text resolut multipl frame type techniqu sort movement frame object requir test xerox xerox figur text block font size approxim binar manual pick threshold output commerci ocr softwar text readabl world figur comparison ocr result interpol estim movement accuraci text static move integ pixel influenc imag interpol improv resolut directli experi encourag result chose text block extract video frame hold interpol shannon interpol imag resolut binar scheme ocr softwar test group imag ocr accuraci origin imag rise hold interpol shannon interpol detail descript topic scope paper found index retriev digit video ultim goal build index system digit video unlik retriev deal full video index base automat extract text interest problem expect miss incorrect charact word poor ocr perform exact match word approxim word match exact word match user submit word consid match problem text digit video lack semant breadth method deal semant latent semant index extens train data similar characterist construct semant dictionari great deal work make impract specif build semant dictionari relat specif topic user submit video frame return put semant dictionari activ investig topic report research result futur present system detect track text digit video automat hybrid network base method detect text region track modul imag match find initi stabil refin match posit system detect graphic text scene text font size track text undergo complex motion discuss ocr index problem video imag current focu make recogn text build video index retriev system experi conduct suggest text enhanc reason ocr result step deal poor ocr result extend semant breadth refer jk motion articul elast motion comput vision imag track medic data deform parametr model page eb baum haussler size net valid neural ellipt head track intens gradient color histogram page black jepson robust match track articul object page black yacoob track recogn rigid facial motion local parametr model imag motion page bregler learn recogn human dynam video sequenc page bs textur segment neural network neural network signal page prentic tm cover geometr statist properti system linear inequ applic pattern recognit ieee tran electron xuan automat recognit provinc plate move vehicl page br hunt imag process neural network recognit cartograph area featur pattern ds doermann video perform evalu technic colleg chellappa document page segment soft integr ieee tran terzopoulo track head mobil object video frame rate page dh folei sampl featur size ieee tran inform hh track occlud vehicl traffic scene page fukunaga introduct statist pattern recognit academ davi track human approach imag understand page yoshida flexibl algorithm book sort system ieee tran guyon applic neural network charact recognit intern journal pattern artifici gd hager pn belhumeur effici region track parametr model geometri illumin ieee tran sinha integr word level knowledg text recognit page rc dc pd system robot harvest machin vision voic signal process techniqu speech recognit noisi environ signal ak jain text segment gabor filter automat document process machin vision ak jain yu automat text locat imag video frame page jone la palmer spatial structur simpl recept field striat cortex cat journal kim effici automat text locat method index structur video journal visual commun imag sk dw hj kim recognit vehicl plate genet algorithm base segment page lee automat extract charact complex scene imag intern journal pattern artifici kim automat recognit car plate color imag process page jf project li ds doermann automat identif text digit video kei frame page ds text extract recognit digit video page tf li ss yu charact recognit probabl distribut featur intern journal pattern artifici automat text recognit digit video acm page multiresolut approxim wavelet orthonorm base tran amer math theori multiresolut signal wavelet ieee tran bs manjunath chellappa unifi approach boundari textur contour ieee tran neural nayar visual learn recognit object page recogn word singl symbol page recogn charact scene imag ieee tran neural network jointli voic featur mouth shape imag pattern de robust method road sign detect recognit imag vision rehg kanad visual track high dof articul applic human hand track page rehg witkin visual track deform model ieee intl conf robot page learn dynam complex motion imag sequenc page sk jm ce jc neural network automat target recognit neural recognit human movement imag sequenc imag dw sk featur select perceptron journal neural network shi good featur track page duncan integr approach track deform displac veloc constraint page boll automat text extract video annot retriev page sung poggio learn human face detect technic ai memo paper pentland track human bodi ieee tran em automat text detect recognit imag understand page ak address block locat complex mail piec page ak jain locat text complex color imag pattern zhou extract text www imag page find text color imag document recognit page 
Agents/1276 1 workshop design person synthet actor june ed robert trappl sort control system aaron sloman school comput scienc birmingham england abstract paper outlin methodolog studi mind part broad disciplin artifici intellig framework architectur requir mind preliminari suggest made mechan underli person descript scenario birmingham framework research problem combin idea work synthet agent inhabit virtual realiti environ person belong agent work ai address cognit aspect design intellig vision form problem learn concept natur languag motor control tini subset ai research concern motiv thing regard relev person partli inspir simon semin late collabor colleagu address issu intim bound deep difficult problem human mind dont expect answer found reason make begin requir design agent specif cognit mechan difficult compon underli implement mechan combin design temporarili architectur kind research detail mechan hope gradual make progress complet realist design theori order make clear framework ill start make gener comment scope methodolog ai ill describ scenario explor architectur account aspect emot person ill sketch preliminari idea explanatori offer complet high level overview famili architectur end includ edit transcript tape record discuss present workshop vienna june remov common misunderstand includ section goal part reason peopl ai term narrow offer altern vision ai broad studi person accommod identifi unsolv problem weak current research identifi aim approach defin aim includ articul larger set goal fit ai unawar lot repeat definit ai origin produc ai recent recent ai ai book claim ai research superset scienc ai peopl tend specialist confer read specialist lose gener vision extern comment ai simpli find underli theme collect idea gener superset aim ai claim previou gener studi modifi control ffl natur ffl actual evolv includ studi individu societi social system add constraint design relat cooper conflict creation understand option differ includ cognit scienc subset ai studi human anim system ai defin number thought ill ai gener studi behav system search algorithm design ai system equal surpass human wed understand aspect design good altern good understand involv design relat design involv aspect implement essenti short understand space region design space properti class design mechan understand good good collect requir design fit question design level issu gener context understand complex cohen stewart terminolog slightli gener word refer collect requir work engin requir specif design fit nich nich physic thing geograph region branch tree nich perceiv thing perceiv thing inform softwar internet occupi comput nich portion world physic locat instanti nich time bee flower locat nich meet function requir defin part nich specif nich type agent ontolog act dynam ontolog event process mean environ type action requir collect task set actual task dynam chang includ depend nich task includ find consum food find learn social fit social constraint law physic legal system rule mean achiev set convent commun nich made constrain ad addit sort food sort agent aid portion world instanti set descript instanti nich simultan nich determin explicitli advanc human engin guid design work nich implicit collect evolutionari pressur oper class organ human design thing complementari role plug compil natur occur implicit nich complement ecolog collect interact chang dynam aspect chang chang climat alter requir discoveri creation suitabl nest chang chang amount kind knowledg alter requir individu collabor competit societi complex societi systemat social role requir individu type person requir extrem differ reproduct role concret physic instanc design abstract determin class design capabl instanti specif design narrow instanc talk impli process discoveri creation design noun design product argu approach defend genet algorithm neural net approach lost search design space true road follow parallel share learn approach empir theoret theoret intuit vagu formal make logic mathemat includ creation studi instanc design instanc creat design implement work part process understand top level goal investig implement common partli analog role thought experi physic design includ specif virtual machin design specifi complex structur interact includ structur featur behaviour capabl compon form causal power function role system differ design includ wai refin common gener start gener mechan appli specif produc parser differ due implement lower level program compil compil machin implement machin architectur physic technolog case design implement lower level mechan case design virtual instanc virtual machin occur physic instanc physic technolog implement comput sparc architectur process biolog evolut mechan explor design space make level result previou reproduct chemic structur neural mechan structur imposs difficult creat instanc design part design includ product process point stress cohen stewart notion dna fulli determin develop ignor role plai mechan dna human design capabl enhanc develop design manufactur tool close design set gener design product mechan nich determin kind design requir enabl facilit specif design scope vari implement detail thing dont understand class design neutral kind high level design intellig agent neutral compon implement collect neural network chemic collect symbol manipul mixtur mechan dont high level aspect design human mind neutral implement human brain abound answer question long term object ai defin relat question extent complex behaviour capabl explicitli built mechan capabl implement capabl directli process learn adapt physic kind directli assembl fulli form adult human brain inform normal adult function case system brain learnt great deal part requir earli person set motiv capabl capabl drive learn process understand give synthet agent person understand sort nich make person relev requir agent fit sort design capabl meet requir iv show complex question studi gener claim constitut scope ai involv found ai work studi studi collect requir constraint agent collect nich understand dimens nich dynam interact nich system made exampl ai includ studi requir nich space design space map design space nich space figur map design space nich space nich set requir map vari kind good kind trajectori design space nich space learn analysi sort perceptu task marr sloman marr comput level analysi paper colleagu concern requir motiv mechan human system sloman sloman sloman beaudoin studi includ analysi comparison design possibl level includ high level low level form type long term short term type perceptu type learn found discuss research paper compar work author rare find discuss design requir complet agent studi map map correspond design class set requir region altern fit nich perfectli style width arrow figur meant kind match map uniqu criterion optim satisfi nich simpl numer measur good fit design nich design nich design catch wider rang prei run good distinguish perceptu limit full analysi varieti map requir analysi logic complex concept peopl understand handl multipl partial order concern type comparison analysi dimens level abstract involv nich design standpoint disciplin studi trajectori concern process extend time complex system interact environ extern behaviour case involv intern behaviour compress chang stress bounc intern behaviour inform includ process develop develop normal infant birth produc capabl system extern agent redesign trajectori design space meet set requir satisfi constraint result trajectori nich space rememb talk movement point movement sort fuzzi explor learn system concern studi mechan move design space nich partli result interact environ trajectori depend environ point view nich part larger system interest question investig sort trajectori design featur make direct implement learn develop turn trajectori design space simpli imposs achiev evolutionari process oper gene pool distribut collect individu physic chemistri imposs individu transform eleph human transform logic organ includ potenti drastic produc gene pool type transform inform process capabl individu imposs born learn understand human born human human share enorm biolog heritag trajectori design space nich space gene pool individu agent turn imposs form conceptu develop exhibit histori human scienc cultur occur matter long individu live mixtur interleav individu develop social learn requir produc transit lead ancient greek scienc quantum physic case process necessarili involv mistaken cognit state enter part environ gener learn avoid state turn true evolut aesthet art moral type person topic studi gener head explor analysi trajectori design space nich space figur rough indic sort thing surfac suggest uniqu level analysi design nich case discontinu design space nich space point import ai theori evolut space discontinu design includ continu variat discontinu region design space smoothli vari chang electr fuel size involv small larg chang wheel inform process lot discontinu design remov condit branch discontinu put back half arbitrari fraction discontinu space capabl continu integ real point feedforward neural net thought larg collect condit action activ condit share action share weight determin influenc rule case influenc vari continu variat condit branch ordinari softwar convent approxim continu variat ad gradual chang probabl condit trigger differ design condit plai design condit discontinu structur chang research topic structur design space find kind discontinu complex design smooth discontinu reason climb search good evolutionari iv learnt colleagu irreduc discontinu understand relat trajectori nich space evolutionari mechan capabl produc discontinu acknowledg evolut requir discontinu chang discontinu ai philosophi hope discuss design nich space trajectori make clear limit conceiv ai search class design meet set requir practic deep studi mind design architectur part broader find sort architectur meet set area design space correspond area nich space enabl philosoph question replac sort mind sort mechan evolutionari learn process bring older philosoph mind suggest answer form design thing kind design kind mind differ similar design relat nich suffici condit order pursu studi set techniqu conceptu tool ill describ requir studi design space nich space languag describ nich work engin develop formal express requir similarli work biologist compar nich relat organ suspect bare scratch surfac requir satisfactori languag account fact nich physic environ nich type agent part design correspond part nich light fuel deliveri system steer mechan car part total design meet decomposit light system includ control power share compon batteri part occupi nich simultan aspect defin featur agent agent agent individu quickli live environ thing chang rapidli abil solv problem rapidli deep standard answer pattern recognit languag nich evolv theori design evolv languag formul design import specif design behav system concern call level newel level dennett stanc level descript level concern design part dennett involv semant content inform level inform process ration newel level dennett stanc evolut design produc system work situat blindli follow rule agent part larger societi ration viewpoint larger system ration viewpoint agent import understand featur motiv person agent concern reproduct care offspr human person human point view individu cost reproduct care young high highli women languag describ map describ map region design space region nich space design evalu rel gener simpl numer evalu includ descript possibl multipl coexist partial order case relat evolutionari effect promot surviv reproduct collect kind fit relev gener determin uniqu design solut nich shown biolog includ potenti futur trajectori lead fit nich individu develop success evolutionari stage cry current content ai book journal expect increasingli import peopl understand issu grasp shallow narrow current ai swing fashion mechan architectur resourc method explor agent design commonplac peopl make unnecessarili limit assumpt mechan methodolog ai ai mechan connectionist neural chemic restrict ai class mechan restrict physic experi mathemat newton ai commit class commit languag part aim ai find formal suit fashion recent year critic problem simul work carefulli chosen toi problem part conquer research requir scienc control simplif hard scienc complic robot ey arm necessarili tackl deep frustrat student found import sort simplif studi bate al call shallow function distinct simplifi find thing high level featur design space nich space build system discov lack qualiti learn architectur dont studi dont work import contribut fuller appreci altern design broad shallow approach perman commit clarifi requir progress deepen design case incorpor learnt narrow deep approach case draw attent miss studi visual percept assum sole function vision produc inform structur motion object sloman expand object explor scientif concern practic engin applic scientif object understand human capabl understand anim understand space design relat nich understand sort trajectori design space nich space condit ai work produc analysi region nich space altern design suffic engin object engin object design machin subset human anim design machin human anim machin compon machin make machin suit interact human depend learn practic object develop educ strategi improv human learn work understand wai human mind brain wrong design therapi easili fix dont work design form toi depend scientif engin requir progress product constant featur ai research design tool includ program languag form process explor implement design cognit affect project cognit affect project birmingham concern issu sever limit resourc forc concentr main effort tini subset task project partial requir preliminari partial specif type architectur explain human preliminari attempt map architectur phenomenolog emot sloman work includ ffl collect abil handl multipl independ sourc physiolog individu tast abil cope concurr process plan abil cope limit capabl level process ffl explor varieti design design mechan manag process motiv gener reactiv mechan chang mechan mechan control manag process aspect control hierarchi accommod long term short term chang ffl produc idea influenc problem involv control attent ffl produc interact demonstr teach primit implement interest initi deepen process work implement reveal inadequaci long run program present import role implement inevit requir contribut inform ftp affect minder scenario order focu investig provid possibl implement restrict develop specif extend scenario studi process interest sloman beaudoin scenario involv simpl collect robot roam interconnect mobil camera mobil hand minder kind troubl rescu develop point leav type problem babi encount includ fall die batteri run recharg point charg die babi turn tendenc damag babi damag babi medic centr repair damag great babi die scenario modifi wai babi move random act goal sort intellig determin nich minder scenario fix set aliv till readi steadi stream incom minder task rate matur surviv babi discharg minder auditori sound troubl trigger visual investig predat make task initi minder movabl movabl camera restrict view limit room minder main task design big task camera room time move cycl control attent process driven current goal altern complex scenario investig simpler agent context evolutionari led poli chose simpl complic motor interest interest control environ design give mind hard control problem work add complex bodi auditori complex previou refer minder part manipul richer visual similarli domain expand enorm complic problem present main focu interest time expect collabor research group implement richer world develop introduc problem commun plan act babi intellig wider rang learn mind possibl investig colleagu call simpl mind task made difficult ffl chang number babi ffl alter rel speed babi move minder bodili part ffl alter rel speed process occur environ process kind occur perceptu gener evalu rel import urgenc import concern requir relat limit process speed high level process constrain design architectur cope asynchron intern extern event simon sloman sloman beaudoin wright al minder collect sort intern parallel specif partli overlap sort thing workshop david bryan loyal intern process includ thing ffl realiz goal ffl decid goal ffl evalu goal likelihood success ffl decid adopt reject goal ffl decid act ffl decid achiev select exist plan ffl detect resolv conflict goal plan ffl carri plan monitor progress task turn unexpect complex turn includ urgenc goal measur time left urgenc function length delai cost benefit urgenc benefit time season plant turn fix sequenc manag task simpl flow chart adequ goal evalu made adopt prior case partial plan requir evalu cost benefit potenti process unpredict amount time problem simplifi minder infinit speed assumpt agent reason process subset cognit amount parallel limit limit parallel observ empir inform found number design factor explain expect amount concurr manag process limit reason appendix sim agent toolkit poli design support explor interact agent rich intern architectur vari rel process speed object agent compon agent toolkit make easi chang speed insid mind simul environ conjectur high level process occur necessarili ant limit speed amount extra level requir call defin process goal involv manag task definit dont mechan process detect plan process long urgent goal decid switch manag process carri partial plan hope plan complet detect rate occurr problem high switch manag task prevent progress lead threshold goal inform dynam attent filter architectur sketch broad architectur agent ai research work work narrow deep concern languag understand main concern put mechan oz group carnegi mellon chosen concentr architectur initi shallow combin sort function research gradual simultan involv complex environ make ad depth ad compon architectur older part human brain share ignor architectur close model live organ studi contribut gener explor nich design space relationship give full account work ill main assumpt sort architectur studi altern automat process manag process clear rout human brain sensor percept action rout share involv form cognit uniqu restrict small subset anim evolv process older rout involv automat process thought essenti larg collect system implement neural permit lot parallel propag activ network process condit action occur action intern trigger exampl includ perceptu postur control process trigger includ intern percept thing temperatur damag food bodi state process trigger process gener environ reflex train control intern gener desir drive learn mechan attent manag process rout percept includ process automat trigger condit intern extern action occur action automat happen altern possibl consid explicitli plan choic made elabor temporari structur creat evalu part process addit arbitrarili complex set store inform access deriv combin inform gener process involv combinatori attempt find combin action enabl problem solv task achiev select complet fragment solut requir choic explicitli construct select made mechan form implement attent manag process automat mechan work automat start function differ manag process automat process kind sort control mechan neural net provid mechan map input space output space fix defin metric automat problem requir creation structur vari complex distinct manag process automat process crude figur partli luc beaudoin ian distinct sharp defin good concept distinct good theori deep concept definit theori shallow start oper definit measur reflex environ architectur motiv agent variabl threshold motiv activ feedback attent filter automat process percept action reflect manag process process action percept figur intellig agent architectur layer involv rout percept action evolv earli share organ common harder terminolog part process theori build concept format featur manag process human resourc unlik automat process automat process portion brain oper manag process share slow parallel explain appendix kind mental task perform count count backward sing song silent altern thing hand talk amount parallel speed manag process interrupt filter mechan prevent disturb manag current task import fail temporarili disrupt goal automat gener automat act part concurr manag task requir plan handl manag system urgent requir interrupt suspend abort activ interrupt limit potenti damag interrupt filter mechan dynam vari threshold requir automat mechan assign measur goal determin abil filter call insist goal insist featur goal inform structur control process process involv motiv sort process involv motiv motiv inform structur potenti initi modifi intern extern directli case innat learnt cognit result process explicit evalu accept format plan execut intern behaviour concern synthet agent person requir ffl motiv gener set ffl mechan suppress protect manag process ffl manag motiv includ ffl assess motiv likelihood cost urgenc ffl adopt form intent ffl condit execut motiv ffl decid execut motiv ffl predict effect occur part evalu plan process ffl assign measur influenc abil motiv maintain control gain control ffl detect conflict motiv ffl detect mutual support motiv ffl set threshold manag interrupt filter ffl termin motiv explicit termin decai ffl detect relev event exist motiv ffl process control manag process decid ffl execut high level manag ffl improv extend improv method assign assess urgenc case ffl extend develop reflex ffl global switch mood optimist turn pessimist cautiou repres motiv structur order engag kind motiv rich structur includ specif featur vari defin motiv fix semant denot state true motiv attitud motiv explicit reason indic current belief ps import base assess consequ descriptor heurist comput determin interrupt capabl correspond estim import urgenc attract attent intens determin motiv attend act continu favour consid motiv kind momentum possibl plan set plan achiev motiv commit statu dynam state defer manag state current relev manag process current robot softwar motiv simpler structur explor possibl varieti type motiv structur requir differ motiv manag process execut process individu differ human explor space show deepen visual percept figur wai mislead suggest compon architectur simpl structur box concern percept complex figur attempt year model visual percept suggest visual system sort complex figur requir percept matter recogn pattern sensori input level analysi requir case speech acoust signal level interpret concern word semant speaker intend achiev call partial view visual architectur form histogram give global inform visibl surfac descript intermedi imag featur bodi imag scene object scene centr descript causal event level trigger motiv emot process control subsystem link postur motiv figur sketch visual case vision requir structur retin imag optic arrai find structur environ properti relat object includ causal relat potenti gibson refer surfac flat sit support cliff potenti danger intern complex varieti task visual process link sourc inform visual imag instanc visual process influenc input current sensor heard gener knowledg type thing form descript spatial structur marr visual mechan output sort includ postur control motiv gener cross link transmit control transmit inform content intermedi environ vision primari claim myriad rout includ rout automat process mechan serv includ control heart sexual sort attract doubt thing dont visual input modul postur aspect partli result physic person height rel doorwai individu aspect due earli childhood easili parent lead develop retir postur control state intellig agent global neural physic event effect semant content short term long term hard slow effect indirect skill semant control state prefer mood emot intent plan figur control state text visual process trigger remind interact influenc produc tendenc parallel common postur control signal detail sort process occur cross link occur depend individu form part uniqu person architect build differ occup percept spider affect person wai person featur high level control distribut system isnt thing distribut person distribut reason hard chang control state intellig agent figur impressionist diagram intend suggest dynam aspect exist level vari type chang dark circl repres event circl repres control long short specif global event fall thought control state subject influenc top influenc longer gener control includ aspect person sudden intens pain hand produc specif short term movement sourc pain long intens pain sort complex includ surgic read differ effect produc similar mental state found mental event endur state semant content state modul process local amount prior action modul quantit speed slow structur switch strategi languag arrow repres differ strength differ determinist probabilist rout system causal influenc link event endur modifi state distinct automat attent process made level diagram circl lower level diagram repres control state influenc event fairli typic readili changeabl moment annoi walk higher system thing indirect influenc capabl broader rang instanc tast food polit attitud influenc event disposit lower system creation motiv control state disposit exist produc actual chang event condit main theme brittl glass capabl determin persist long time produc glass hard similarli mental state produc effect condit level state gener rang potenti specif state control current postur specif prefer movi directli influenc specif item behaviour board train movi heard make close poster film show make librari read book high level control state evolv instanc tast film develop time manifest high level disposit mental state chang inform produc behaviour produc influenc earlier chang manifest long fact disposit activ condit thing suppress influenc exist intern lead extern sign peopl lot state manifest reveal aspect sort control exist express current import effect social complex social feedback loop differ level control hierarchi correspond direct differ manag process automat process depict figur differ level sort discuss section occur manag process automat process understand issu requir explor design space type control architectur problem defin manifest person varieti emot state individu capabl discuss fact word wide varieti definit peopl claim studi emot extrem peopl write motiv strong motiv person strong desir promot emot id call motiv kind extrem view defin term mechan system attent concept emot appli human anim share ancient evolutionari histori defin term extern behaviour corner mouth aggress postur defin term thing boundari intern extern muscular propriocept sensori skin concentr subclass emot state common human occur anim state deepli involv high level cognit process social excit pride kind emot poet write state partial loss control high level cognit process includ common word avoid technic definit project word label phenomena word intend reson concept perturb emot state heard woman interview radio talk son kill road accid year talk state muscl facial talk state mind theme extent unabl control mental process constantli drawn back prevent room children move inabl control meant pai attent normal task carri properli found inabl life made affect peopl thing includ children made depress felt inabl cope life end emot pile top possibl sort mental state occur kind consequ type architectur talk theori emot claim occur time link architectur state discuss fulli al case suspect peopl experienc state person made fool public remain concern time simpli happen frijda frijda stronger hold attent put mind drawn back dont anger common stop achiev goal return stop desir sort perturb state talk aspect anger addit find partli control desir back put mind continu draw attent thing disposit disposit temporarili swamp power child sudden danger suspect state occur anim share state process occur call part sort architectur make control thought process rat lose control thought process perturb state possibl perturb state inher sort architectur architectur evolv order produc perturb state similarli possibl thrash inher comput oper intend design possibl emerg function aspect design case manag mechan parallel protect kind filter mechan iv suggest motiv inform assign level insist determin abil filter interrupt manag process insist level assign automat process prevent high level insist assign goal conflict high level object attent frequent interrupt held item assign high insist level inconsist goal strategi select agent partli lost control thought process full account perturb state show disposit state persist temporarili insist accept thought goal grab hold attent wright al explor nich space design space relationship includ explor infer natur artifici attempt kind behaviour capabl environ understand human person context studi design agent nich studi prerequisit construct synthet agent aspect human emot motiv synthet agent paper show featur explain basi underli inform process architectur design requir implic studi ffl ffl ffl ffl ffl view fit scheme thing acknowledg work support uk joint council renaiss grant birmingham grate robert trappl colleagu austrian research institut artifici intellig prepar transcript talk discuss luc ian tim christian chri past present member cognit affect project birmingham addit paper group found ftp web page discuss edit transcript discuss lead talk part tape unclear edit transcript check access origin audio tape wrong ambigu incomplet portion transcript move portion discuss back main relev metric success design metric success design dont notion success easili measur babi minder manag aliv period time exist simpl vari rel speed movement babi movement minder intern process minder find imposs babi set babi depend initi geograph distribut babi random element sort full implement includ make extern event gener goal faster handl manag introduct process improv success babi aliv physic resourc limit point clarif emerg claim manag process resourc limit contrast automat process point reason parallel limit part architectur physic hand physic resourc limit distribut case robot thing fewer physic limit parallel talk physic resourc limit talk limit inform process resourc interest question fact evolv resourc good design reason good question discov fact human good design reason limit parallel fact learn thing solv complex credit assign decid combin good bad thing happen unexpectedli number combin exponenti function number rapidli effect delai type limit memori make highli parallel associ question time answer type limit concern requir build temporari structur unbound plan problem solv due singl coordin process level reduc risk independ mutual inconsist chapter design reason accid sophist agent limit number task simultan automat process limit parallel dont access common inform dont sophist dont build temporari structur unknown complex workspac notic automat process build complex visual languag understand process effect form train alloc dedic portion automat mechan task run parallel higher level process learn music circumst mechan minim mutual mention work resourc limit system differ architectur present good wai studi mental find organ architectur type discuss suspect find organ share call automat process impress speci includ attent manag process process control bird perform complex task automat mechan respond trigger produc stereotyp unlik human builder explicitli consid altern altern altern construct sequenc act meta manag process human approach defin research programm biologist filter implement bit intern percept intern prefer david mention filter system doesnt claim goal interrupt manag process total open mind implement implement neural dont explicit activ condit oper excit inhibitori link altern separ mechan explicit threshold set higher level process import point sort function total architectur explain requir kind nich requir kind design give nich minder simpl number featur chosen drive research world number thing happen capabl time gener goal system give system make select plan complet back simpl comput interfac typic window asynchron perceptu monitor concurr insid problem requir form part nich requir speed thing happen environ make intern resourc limit show intern architectur round problem resourc limit mean system deal goal immedi investig fulli potenti consequ cost benefit achiev decid adopt process import current plan process problem manag mechan support unlimit parallel speed thing happen slowli environ intern process run complet order opportun miss requir environ unpredict mean system decid advanc interrog environ build plan asynchron concurr perceptu process gap transcript due tape chang slightli sophist simul minder awar cry decid find problem contrast simplest version minder problem manag task decid act problem achiev babi low charg recov recharg pick move babi close current domain minder small collect requir subject babi troubl requir make design problem ad requir instanc add requir cooper order thing sensibl minder babi minder shouldnt babi nearest minder comparison schedul system kind problem solv comput dedic person problem schedul comput infinit infinit design function optim condit perform defin notion optim assum minim architectur thing make work perfectli open decid explor sort assumpt sort talk run genet algorithm evolv natur varieti situat find collect solut work wai thing map nich space design slide style isnt simpl notion design fit fit nich wai fit work normal slight chang circumst make design dont work normal work abnorm condit similarli design speed costli easier build varieti kind issu suspect dont understand issu explor partial answer question altern solut predict friendli understand environ understand implic deal unpredict environ environ minder minder babi lot constraint wont room partli predict environ partli predict part result environ make live persist featur concept cognit friendli environ dimens cognit friendli avail detail inform didnt electromagnet radiat bounc surfac thing detector retina tune environ cognit friendli type cognit friendli concern amount inform complex process requir discrimin physic structur object importantli behaviour hard differenti tree produc fruit produc fruit count leav tree find number odd cognit unfriendli thing evolv sort case occur environ total unpredict level manag question process defin level manag requir run step feel suffici independ nich characterist perform explor larg number nich space sort priori empir fact abil point work suffici gener doesnt solv problem nich peopl encount motiv level explor nich lead addit aspect architectur addit aspect requir concurr manag claim dont reason explain limit concurr rule coexist process involv build robot free put mechan find total sort mechan discuss control system requir complex autom factori sort architectur regard open nich wai wai requir architectur nich success solut sort mind filter identifi process identifi process control manag explan human emot havent relat emot talk total control thought process link fact attent mean potenti user resourc automat resourc imagin design start attend problem divert problem solv disastr case import urgent turn motiv captur attent manag process motiv manag process limit parallel task urgent import requir full attent detail dont brain surgeon oper brain distract sound thought holidai spain similarli give intric instruct listen miss crucial inform possibl interrupt manag prevent notion dynam variabl filter mechan vari open question origin start idea simpl threshold luc phd thesi suggest simpl numer filter qualit divert attent kitchen type level threshold determin process determin current manag task easili interrupt insist level automat process gener potenti distractor assign insist valu fulli resourc filter mechan protect disrupt insist assign mechan dirti heurist give approxim estim import base learnt pattern thing type interrupt occur insist assign occur sort filter threshold set kind circumst person vari thing call human emot state case continu grab dont mind kind pleasur partli control thought good thing peopl put situat partli freeli kind strang thing peopl claim aspect emot state call manag process partli occur context architectur suffici rich support possibl manag process occas control lost requir process evalu control manag process dont architectur make process architectur good current network server architectur worth back babi pretti convinc comment formul network server model system resourc reconfigur network complic basic serv custom system run situat period collect statist defin heurist local give static state behaviour dont static state situat emot content find violat kind formal model peopl kind situat convinc approach ai network lot distribut control rout talk talk serv lot comput question sourc motiv network server thing request babi class request arriv asynchron vari import urgenc finit death injuri problem domain express idea million analog problem domain question plan network depend problem case perform simpli numer statist maxim likelihood success oper claim human find thing difficult partli explain find optim engin solut problem sort architectur sort mechan wouldnt architectur planner collect wai figur basi distribut babi arriv distribut babi death time real eventu learn control fast heurist decid learn kind eventu build system design evolut tune adapt process plai import role automat process share anim flexibl deal circumst chang statist adapt process construct make complex form complex system concept nich human architectur fit complex typic network server present complic kind system point kind schedul optim condit matter extern problem architectur solv problem differ faster process kind claim uniqu distinct solv creat optim goal make work make turn easi lot wai make environ real human life regard fashion intellig agent explicit plan construct problem solv prior perform action workshop explor altern date mode travel creat includ decid airport coach car time interf plan kind nich defin research problem requir minder perform plan task compet engin find solut simpl domain made plan rule artifici domain evalu evalu isnt compar human kind metric artifici human circumst optim havent optim sort thing develop intuit made person dont direct answer thing person peopl class find framework didnt make miss architectur reason make reason human situat make made term time pressur happen start make neg point difficulti make comparison shouldnt distract ill give posit answer lower direct wai evalu system human drive simul compar behaviour produc softwar system indirect long time indirect wai framework gener explan thing psychologist social worker teacher concern produc mention paper emot al trigger read report experi intens long written love friend di long ill paper made shot show phenomena architectur process forc pai attent thing miss includ support experi pleasur kind defin meant control thought process led develop implement comparison model detail phenomenolog report sort evalu show paper expert clinic psychologist pleasur paper relev read editor journal psycholog immedi publish prove idea correct partial initi indic make progress direct regard kind indirect doesnt idea idea peopl havent turn scienc prove correct gradual closer closer accur theori test run simul point event happen simul find behaviour correl peopl observ behaviour dont correl peopl find explan term featur architectur interest explain problem address enorm complex solv lifetim current limit current understand architectur mechan formal time dont expect main form test studi actual behaviour work implement idea earlier mere task plan implement make discov gap make understand problem send back draw board extend requir specif nich aspect design dont implement produc implement peopl pai attent idea show work prove empti drive develop real unlik theori hope eventu produc nice teach tool psycholog student dont run run guess explor understand design space run bit extrem impact behaviour havent thought back understand person dont long line experi compar implement human behaviour advanc larg discrep implement futur peopl variabl regard import deep analysi effect differ design situat affect speed correct move higher level understand test understand test consist idea implement dont simpli run lot test program formul theoret predict architectur capabl run program relat theoret explan understood result architectur model person architectur start model variat chang manag complet real biolog social structur lot gener system gener differ individu start support diverg function societi differ amplifi differ individu grow vienna grow african minor variat expect kind person accommod gener sort motiv strategi evalu compar wai assign prioriti differ automat part main focu research long explor type genet differ human find wai implement interest make peopl happi airlin pilot bu kind life prefer claim genet determin depend extent individu absorb inform strategi gener perceptu categori process grow cultur regard chang acquisit type abil link compon rememb person thing larg collect inform control structur distribut system long term research suspect explain gener sort architectur interest individu variat term kind wai thing agent react situat wont varieti toi domain support richer scenario support individu variat sort human includ environ develop larg set detail problem individu solv structur environ produc long term affect futur process behaviour full studi person requir investig rang wai individu genet share common gener birth implement dont tini subset refer bate loyal broad paper present aaai spring symposium integr intellig sigart aug pp lp beaudoin studi motiv process prospect artifici ed partridg io goal process autonom phd school comput cohen stewart penguin emot cambridg press make robot consciou mental aaai spring symposium repres mental state access vision societi william ortoni gl clore cognit structur cambridg refer featur guid reason opportun fourteenth annual confer cognit scienc societi lawrenc erlbaum associ concept hutchinson motiv emot control reprint model yale comput revolut scienc model harvest press human sloman robot proc int joint conf vancouv interact philosophi role intuit reason proc int joint conf reprint artifici enabl machin proc int joint conf ai lo motiv mechan cognit emot reprint philosophi artifici design visual system comput model journal theoret note aisb theori commun ortoni slack commun artifici intellig theoret appli germani prospect ai gener scienc partridg prospect artifici io amsterdam mind control philosophi cognit cambridg sloman explor design proc european confer sloman semant intellig control philosoph transact royal physic scienc explor design space nich proc scandinavian conf io amsterdam poli sim toolkit explor agent intellig agent vol ii ed mike milind springerverlag pp present workshop agent august cognit scienc technic summari attent affect affect file project beaudoin lp analysi emot philosophi psychiatri psycholog 
ML/1143 6 guid oper genet algorithm base optim khale comput scienc depart rutger nj usa algorithm extens domain mean global optim simpl reliabl manner chanc global optima method converg local ga tendenc moder close optima small number iter close ga larg number optim close local optima small number iter paper describ crossov oper design endow ga abil comput gradient sacrif global optim oper work guidanc member ga popul select direct explor empir result engin design domain demonstr oper significantli improv steadi state error ga optim introduct genet algorithm search algorithm mimic behavior natur select ga attempt find solut problem maximum gener collect potenti solut problem mutat recombin solut gener current set potenti solut process continu accept good solut found ga advantag search includ abil deal qualit type continu variabl discret quantiz variabl variabl domain continu domain local search method algorithm tend advantag abil extrem close local optima small number iter ga larg number iter close optima ga higher probabl reach global optimum method classic approach solv dilemma ga optim run local search method complet optim approach problem problem decid ga reach good region thought basin attract global optimum case highli multimod region small run local search method time select point final ga popul problem local search method suffer paper present crossov call capabl improv steadi state error ga optim distanc point found global optimum search continu gradient inform method demonstr merit oper ga optim realist engin design domain domain explor section design repres number continu design potenti solut vector vector space determin qualiti point involv simul comput relev physic properti artifact summar singl measur merit simul amount rang fraction dai gradient extrem comput tabl lookup high search space make difficult method high reliabl demonstr util guid crossov set benchmark domain engin design optim studi skolnick remaind paper present detail descript oper present number experi oper realist engin design task conclud paper discuss relat effort futur work ga architectur ga research detail al hirsh individu ga popul repres parametr descript paramet continu interv fit individu base sum proper measur merit comput simul takeoff mass penalti function relev legal limit permiss oper appli element popul select scheme select perform rank actual fit function wide rang fit valu penalti function rank select prevent discov point domin popul steadi state ga model exist point popul replac newli gener point replac strategi replac strategi fit proxim point ga goal select replac point low fit close point introduc ga stop maximum number evalu exhaust popul complet lose practic converg singl point search space ga architectur includ screen modul hirsh hirsh mainten modul turn sm save time prevent ga evalu point close encount bad point learn techniqu maintain prevent ga evalu candid point extrem close evalu point attempt restor oper sever loss detect earli stage optim section describ crossov oper detail guid crossov guid crossov work candid point select ga popul normal select rule call candid candid point select ga popul point ga popul candid quantiti mutual mutual euclidean choic maxim mutual candid point fit current ga popul candid candid swap make candid point higher fit result crossov point line join candid candid select random small region candid result candid candid uniformli distribut random number interv function number iter total number iter gc examin direct form join randomli select candid point point current ga popul direct rank base contribut give object function move end point rel distanc end point direct rank small step direct vicin end point magnitud step ga optim progress guid crossov oper crossov oper ga greedi natur substitut crossov oper fraction time current random choic made iter gc convent crossov oper probabl gc linearli maximum number iter maximum maximum extern paramet default intuit guid crossov endow ga close optimum advantag claim method ga costli comput gradient potenti evalu highdimens space quantiti mutual rank direct serv crude form gradient calcul entail singl addit evalu fit result evalu ga oper appli design problem modern engin design conceptu design superson transport design case point fit candid effort forc point end optim local optimum tabl aircraft paramet optim paramet exhaust converg length exhaust diverg length exhaust extern length exhaust engin size wing area wing aspect ratio taper length effect structur wing sweep design mach angl wing taper ratio fuel width superson missil benchmark engin design domain note merit competit optim approach respect tradit optim method domain demonstr research effect crossov oper ga ga behavior demonstr merit guid crossov ran experi guid crossov guid crossov guid crossov replac heurist crossov oper heurist crossov oper similar principl guid crossov heurist crossov oper work parent fit newborn random select uniformli interv heurist crossov view simpl version guid crossov section discuss result domain superson transport aircraft design domain concern conceptu design superson transport aircraft summar briefli detail al figur show diagram typic airplan automat design softwar system ga attempt find good design mission vari aircraft conceptu design paramet tabl continu rang valu ga evalu candid design multidisciplinari simul current ga goal minim takeoff mass measur merit commonli aircraft industri conceptu design stage takeoff mass sum fuel figur superson transport aircraft design system rough approxim oper cost rough approxim cost build aircraft complet mission simul requir cpu time dec alpha desktop workstat aircraft simul model base implicit explicit assumpt engin approxim numer optim human domain design paramet set correspond aircraft violat assumpt physic realiz simul detect fact reason set constraint introduc safeguard optim process violat penalti function approach incorpor effect constraint violat penalti function ad takeoff mass return simul result sum quantiti optim minim serv fit assign point ga specif penalti function simpli larg constant multipli evalu figur effect guid crossov aircraft design domain sum amount constraint violat violat constraint presenc constraint induc partit search space mutual unevalu point repres design violat model assumpt simul complet simul process produc inform point larg fictiti takeoff mass gener object function evalu point repres aircraft type extent model violat moder simul complet work report constraint violat inform penalti function ad takeoff mass return simul account model violat feasibl simul succe evalu mass point violat occur process penalti function point ten random popul point popul ga proce iter iter denot call gc time heurist crossov replac gc result experi shown figur graph plot measur merit point found number iter curv repres averag run ten start popul solid curv repres averag ten run default paramet includ guid dot curv repres averag run heurist crossov dot curv repres tabl inlet paramet optim param definit initi cone angl final cone angl axial locat throat radial locat throat axial locat end cross section intern lip angl ej height end constant cross section fk height begin constant intern cross section averag run paramet adjust guid crossov figur gc improv steadi state error domain fact hc averag final improv intermedi perform ga search perform run gc gc hc superson missil inlet design domain concern design inlet superson missil summar briefli detail al missil inlet design mix compress inlet mach feet altitud minim manufactur cost inlet techniqu boundari layer bleed variabl geometri perform inlet reli sole aerodynam design rigid extent extern intern contract inlet start throat throat shock train diverg simul domain program call develop unit technolog research center inlet tool al design paramet continu tabl coordin term axial radial posit object optim minim total pressur quantiti commonli measur perform inlet aircraft experi consist ga run guid crossov heurist crossov replac guid start random initi popul domain popul consist random individu popul ga proce iter iter denot call evalu figur effect guid crossov missil inlet design domain tabl benchmark domain domain dim constraint inequ result shown figur plot measur merit pressur function iter number figur demonstr amount improv stage optim realiz guid crossov benchmark domain problem introduc eric phd thesi engin design optim research benchmark recent experi involv domain report skolnick ga packag call numer optim packag call compar domain domain repres sampl origin ran experi domain minim problem properti benchmark summar tabl column tabl show problem number thesi unabl comparison domain unbound variabl evalu figur effect guid crossov benchmark problem evalu figur effect guid crossov benchmark problem column show problem dimens number design variabl fourth column show number inequ equal constraint sixth column show optima problem detail descript domain detail descript experi compar optim domain found paper focu studi effect guid crossov perform experi perform random start popul domain run gc hc replac gc time gc result shown figur figur clear gc enhanc perform domain final perform gc exclud benchmark problem final analysi run setup find feasibl point unabl qualit comparison evalu figur effect guid crossov benchmark problem evalu figur effect guid crossov benchmark problem evalu figur effect guid crossov benchmark problem evalu figur effect guid crossov benchmark problem evalu figur effect guid crossov benchmark problem perform gc stage optim domain final comparison final remark crossov oper continu space search present result demonstr merit oper domain aircraft missil inlet design benchmark engin design domain research effort appli genet algorithm engin design optim problem varieti includ control system design architectur civil engin design al vlsi design mechan design deb aircraft design al effort includ innov ga oper crossov oper contribut lot improv ga optim idea step prove power direct select join line point ga popul line interest idea creat direct combin line plan explor method futur acknowledg aircraft design gene invalu assist research member haym donald keith research partial support nasa grant part project support advanc research project agenc depart contract bibliographi chapman genet structur topolog design complianc topolog simplif journal mechan deb robust optim design techniqu mechan compon design evolutionari algorithm engin page al andrew don smith model knowledg guid design space search fourth intern confer artifici intellig design al john vladimir thorsten genet engin design problem evolutionari algorithm engin page david goldberg genet algorithm machin learn al sobel inlet design analysi final report report kundu ai control system design paradigm design fourth intern confer artifici intellig design jen genet algorithm channel rout vlsi circuit evolutionari zbigniew michalewicz genet algorithm data structur evolut program al nakamura multiobject genet algorithm multidisciplinari design wing platform journal skolnick powel skolnick genet algorithm engin design optim constraint intern confer genet page morgan juli hirsh khale haym hirsh learn improv design optim seventh intern confer genet algorithm morgan hirsh khale haym hirsh learn select design optim artifici intellig analysi al khale haym andrew genet algorithm continu design space search artifici intellig elsevi scienc khale genet algorithm continu design optim technic report depart comput rutger januari phd gener form evolutionari approach evolutionari algorithm engin page eric util nonlinear program algorithm technic purdu phd thesi al don mark khale andrew doyl knight high perform superson missil inlet design autom optim aiaa symposium multidisciplinari analysi optim 
DB/99 3 offer tradeoff aggreg queri replic data chri jennif widom stanford abstract strict consist replic data requir distribut current system permit stale cach copi data valu date queri cach data return answer stale answer queri remot master data return potenti poor perform bridg gap class replic system call replic system give user control tradeoff cach store rang bound current data store stale exact valu user suppli quantit constraint queri answer system automat select combin local cach bound exact master data store remot deliv bound answer consist rang wider comput quickli paper defin architectur replic system cover mechan cach data rang queri present optim algorithm answer queri report perform experi demonstr control tradeoff offer system introduct environ replic inform multipl site permit stale enforc exact consist multipl copi data exact consist perform perspect larg varieti reason outlin distribut applic exact consist simpli requir web gener stale replic master copi page maintain web server stale copi cach web browser web read stale cach data browser significantli perform retriev master copi web server press browser cach copi arbitrarili date stale replic system data view data object oper master data multipl stale cach copi queri cach data typic faster queri master copi oper site work support nation scienc foundat grant nasa grant nation scienc foundat graduat research fellowship run scenario motiv exampl simpl replic system monitor network link thousand comput assum node network track averag traffic level incom network link node administr monitor station analyz statu network collect data period network node link monitor station cach latest traffic level figur obtain node administr queri bottleneck bandwidth path total latenc path averag traffic level minimum traffic level fast link link high bandwidth low link high averag latenc link high administr obtain current answer kind collect data valu relev node time queri pose long affect system requir node constantli send updat valu monitor gener unnecessari paper develop approach replic queri process user control tradeoff answer high perform traffic level figur monitor cach exact node send updat exact move cach rang queri execut cach rang return rang current exact answer administr pose provid constraint indic wide rang toler answer administr sampl peak latenc period critic order decid monei invest upgrad network make administr peak latenc obtain answer millisecond system automat combin cach rang valu retriev node order answer queri quickli tradeoff stale replic system potenti offer user mode queri call queri sourc answer potenti poor perform call queri execut cach data answer quickli answer user left guess perform cach data sourc data current replic system perform combin cach sourc data system figur tradeoff base knowledg data stabil recent cach updat figur illustr tradeoff extrem queri mode discrep extrem point figur lead answer obtain mode obtain place heavi load system lead unaccept delai applic requir level extrem point run administr pose queri quantit constraint find middl ground sacrif sacrif perform address kind replic call replic support monoton tradeoff figur queri accompani custom system answer queri combin cach sourc data optim perform meet constraint extrem point system correspond queri mode defin overview approach addit introduc paper focu specif replic system call queri aggreg numer data convent answer queri outermost aggreg oper singl real defin bound answer call bound pair real valu la ha defin rang ha answer lie quantifi width rang la exact repres unbound constraint constant denot maximum accept rang ha la give bound ha queri requir cooper data sourc cach sourc refresh cach data object current exact sourc send rang call bound cover gener case bound function sourc link latenc bandwidth traffic refresh weight cach cach cach cost figur sampl data network monitor actual stai exce bound sourc immedi send refresh cach store bound data object exact cach assur current master bound cach answer bound valu store comput express term bound small tabl figur show sampl data cach network monitor station section current valu network node weight row figur correspond network link link node link node recal master valu traffic incom link measur store link node monitor station store bound traffic cach bound valu comput bound answer queri bound answer queri aggreg comput cach answer satisfi user answer bound wide data refresh sourc improv assum quantit cost refresh data object cost vari data item base node distanc network path show sampl refresh cost figur system optim algorithm attempt find combin cach bound master valu answer order minim cost refresh constraint offer continu relax constraint queri enabl system reli cach improv perform queri tighten constraint system reli master degrad perform yield answer contribut specif contribut paper defin architectur replic offer user control tradeoff method determin bound comput standard relat aggreg function bound data queri select join present algorithm find set tupl refresh order answer aggreg queri select predic discuss optim algorithm analyz complex case exponenti suggest approxim implement algorithm present initi perform result relat work larg bodi work dedic system improv queri perform give approxim answer earli work area report system precomput sampl give answer statist estim scan input data system scan data bound exact provid statist result previou work similar idea move object system replic scheme cach valu permit deviat master valu bound amount unlik system answer queri combin cach master user control tradeoff bound data object set independ constraint bound set static system administr move object bound set maxim singl metric combin elimin user control tradeoff system support aggreg queri demarc protocol techniqu maintain arithmet constraint distribut system system relat work bound data form arithmet constraint demarc protocol design modifi arithmet constraint system updat bound demarc protocol deal queri bound data aggreg queri select approxim approach produc bound answer time permit select predic evalu tupl approxim deal queri bound data work deal queri fuzzi set bound valu consid infinit fuzzi practic approach fuzzi set approxim exact valu cost relat data model data object undergo lossi compress reduc size read compress version data object full system quickli produc approxim answer queri system perform improv reduc number data object read remot reduc size data diverg cach bound number updat permit master copi data object cach bound data valu bodi work deal inform system inform qualiti iq system quantifi accuraci data granular entir data server bound individu data queri concret knowledg individu data valu form bound answer iq system give bound answer queri data object valu rang consid special case constrain valu constraint null variabl local condit incomplet inform work area consid constrain valu bound approxim exact valu store aggreg queri set uncertain membership due select condit bound consid system architectur architectur system illustr figur data sourc maintain exact data object data cach store bound exact valu sourc valu multipl cach possibl cach bound valu multipl sourc user submit queri queri processor local data constraint answer queri queri processor send refresh request refresh monitor respond bound refresh monitor sourc track bound data object relev cach network monitor applic sourc track small number bound applic sourc provid larg number object multipl case scalabl trigger system great benefit refresh monitor detect data object exce bound send bound cach cach bound data object refresh cost incur gener case object cost practic cost refresh object depend sourc amort refresh cost set discuss section cost algorithm tupl refresh order meet constraint queri minimum cost architectur present paper make simplifi assumpt object insert delet occur regular basi insert delet handl propag immedi cach discuss limit level offer system account time send refresh messag process singl queri assum time refresh bound small introduc refresh monitor user constraint queri queri processor data cach bound sourc data answer refresh request refresh figur system architectur assum refresh occur time individu queri process address issu topic futur work discuss section section discuss detail mechan bound valu refresh section gener bound function function section discuss execut aggreg queri present specif optim algorithm aggreg queri section section present preliminari result aggreg queri join refresh cach bound master copi data object resid singl assum singl real denot cach store rang valu data denot sourc send copi data object cach refresh event time addit send current denot send bound discuss refresh occur reason master data object exce bound store cach current time sourc oblig refresh cach current bound refresh refresh occur queri execut cach requir current exact data object order meet constraint sourc send bound queri bound function time section present simpl approach bound data object pair endpoint gener accur approach bound endpoint bound function time function properti refresh initi initi time figur bound overlaid refresh time bound time refresh width endpoint equal current time advanc past endpoint bound diverg bound time refresh sourc send pair bound function cach replac pair figur illustr bound data object overlaid subsequ algorithm result paper independ bound select bodi paper assum bound function evalu current time write write exact current preliminari work investig bound deduc absenc addit inform updat function expand time time recent refresh determin width chosen interest reader refer appendix detail queri execut bound answer execut queri constraint involv combin data store remot sourc bound data store local cach section describ gener bound aggreg queri present cost model algorithm cach data object refresh answer queri remaind paper assum relat implement data model support aggreg numer valu queri form join address section select predic aggreg standard relat aggreg avg predic predic involv column tabl possibl constant nonneg real constant requir bound answer ha queri satisfi ha la omit implicitli comput bound answer queri execut comput initi bound answer base current cach bound determin constraint met algorithm refresh examin cach copi tabl subset tupl tr refresh sourc tupl tr refresh cach copi tupl refresh recomput bound answer base cach partial refresh copi refresh algorithm ensur answer step satisfi constraint section present detail base specif aggreg queri select predic type aggreg queri address comput bound answer base current cach bound problem correspond step set tupl refresh problem correspond step refresh algorithm optim find cheapest subset tr tupl refresh subset total final answer queri satisfi constraint valu refresh tupl current bound assum cost refresh set tupl sum cost refresh member order optim problem manag simplif ignor amort due batch multipl request sourc recal assum separ refresh cost assign practic tupl sourc incur cost note entir set tr tupl refresh select refresh constraint valu tupl tr approach refresh tupl time sourc comput bound answer refresh stop answer section discuss aggreg select predic section specifi comput bound answer bound data valu type aggreg describ algorithm select refresh set aggreg function assum select predic queri involv column exact valu section assum select predic appli aggreg comput tupl satisfi predic queri select predic involv column bound valu cover section join involv bound valu discuss section comput aggreg column ta cach tabl ta tupl store cach bound comput queri processor option tupl read cach bound refresh obtain master cost refresh final answer aggreg bound ha comput min select predic comput bound min ta ha min lowest minimum occur bottom bound highest minimum occur tupl return section find minimum bandwidth link path queri appli bound min bandwidth tupl figur yield optim set tupl refresh min queri constraint algorithm justif proof optim nontrivi appendix refresh algorithm tr tupl min independ refresh cost tr tupl lower bound minimum upper bound minu constraint index exist upper lower set tr found time index upper bound find min index lower bound find tupl satisfi min run time refresh queri find minimum bandwidth path refresh refresh tupl tupl low min tupl bandwidth turn bound answer max aggreg function symmetr min appendix detail subsequ defin section briefli discuss index rang problem activ work comput sum select predic comput bound sum sum valu ha smallest sum occur valu low largest sum occur valu high run bound sum latenc path data figur problem select optim set tr tupl refresh sum queri constraint attack problem select tupl tr tr observ ha la refresh tupl tr valu contribut bound equal combin constraint ha la give constraint optim object satisfi constraint minim total cost tupl tr observ minim total cost tupl tr maxim total cost tupl tr optim problem formul tr maxim constraint turn problem isomorph knapsack problem state set item weight profit knapsack capac hold set item long total weight goal knapsack problem subset sk item place knapsack maxim total profit knapsack capac sk maxim constraint state problem select refresh tupl bound sum queri knapsack assign sk tr knapsack problem approach solv problem dynam exponenti run time approxim algorithm exist polynomi find solut total profit fraction optim run time algorithm log algorithm refresh adjust paramet algorithm trade run time algorithm qualiti solut special case uniform cost tupl knapsack object profit knapsack problem polynomi algorithm optim answer found object order weight knapsack hold object add tupl tr start smallest bound tupl index exist bound width section algorithm run sublinear time index bound run time algorithm log refresh time total refresh cost approxim paramet total cost figur refresh time refresh cost vari queri total latenc path figur show correspond problem knapsack problem knapsack latenc column tupl exponenti knapsack algorithm find total latenc path tupl capac leav tr bound sum latenc refresh tupl perform experi refresh approxim algorithm quickli find cheap set tupl tr refresh constraint hold implement algorithm ran experi actual stock price vari highli dai high low valu dai bound close refresh cost data object set random number run time measur sun model run suno figur fix constraint vari knapsack approxim order plot refresh time total refresh cost select tupl smaller valu refresh time refresh cost refresh time quadrat refresh cost small gener advantag set close refresh extrem figur fix approxim paramet vari order plot constraint versu perform refresh refresh algorithm concret instanti figur show monoton tradeoff perform system sum valu constraint figur tradeoff refresh comput count select predic select predic comput count amount comput cardin tabl requir insert delet propag immedi data cach cardin cach copi tabl equal cardin master refresh comput avg select predic select predic procedur comput avg aggreg comput count discuss section simpli cardin cach comput bound sum section count produc sum ha count count bound width ha la count comput sum count ha la constraint satisfi run time domin run time refresh section queri section comput averag traffic level entir constraint comput count comput sum count column label figur show knapsack weight assign tupl base cach bound traffic optim knapsack sum comput tupl result bound sum divid count bound avg latenc refresh refresh refresh refresh refresh refresh figur classif tupl select predic modif incorpor select predic select predic involv bound valu present comput bound aggreg result refresh tupl meet constraint complic section present modif algorithm section handl aggreg queri select predic begin introduc techniqu common queri aggreg function present select predic involv column bound valu system partit disjoint tupl possibl satisfi predic current bound data tupl satisfi predic current bound data tupl mean exist valu current bound predic valu predic satisfi process classifi tupl select predic involv column bound valu detail appendix interest aspect filter find tupl express simpl predic bound algorithm comput bound answer tupl refresh examin tupl classif express sql queri optim possibl incorpor special index discuss section exampl remaind section refer figur show classif predic data figur exact valu refresh comput min select predic select predic bound min answer ha min min la tupl satisfi predic turn smallest lower bound tupl satisfi predic form lower bound answer ha tupl satisfi predic turn smallest upper bound tupl satisfi predic form upper bound answer run queri find minimum traffic result data figur classif figur tr tupl min algorithm essenti refresh correct optim reason appendix addit case refresh tupl move tupl contribut actual affect bound answer ha constraint hold refresh run time sublinear index upper lower bound run time queri constraint tr tupl pass select predic low valu min tupl turn pass select bound min max aggreg function symmetr min appendix detail comput sum select predic comput sum presenc select ha la occur tupl neg valu satisfi select predic contribut result ha occur tupl posit valu satisfi predic algorithm similar refresh map problem knapsack problem modif requir ignor tupl tupl set valu idea effect extend bound tupl includ tupl contribut sum contribut knapsack extend bound adjust weight comput count select predic bound answer count aggreg function presenc select predic ha queri section number link latenc figur show classif tupl bound count algorithm base fact ha la refresh tupl remov optim algorithm tr cheapest tupl index algorithm run sublinear time run time requir sort log queri select tr cheapest tupl updat tupl turn bound count comput avg select predic comput bound answer comput bound avg predic present complic comput aggreg count bound longer simpl matter divid endpoint sum bound exact count section comput lower bound start comput averag low endpoint averag low endpoint bound time order point averag comput upper bound avg queri section averag latenc link comput lower start averag low endpoint tupl averag low endpoint tupl obtain lower bound averag latenc stop point averag tupl lower bound appendix formal run time log looser bound avg comput linear time comput sum sum count count algorithm section ha sum count linear algorithm yield notic bound looser bound achiev algorithm tupl refresh complic scenario detail provid appendix give descript algorithm fact bound avg achiev function bound sum linear algorithm section refresh tupl provid bound sum count bound avg function bound sum count meet constraint interact accomplish modifi version algorithm understand choic refresh tupl sum affect bound count algorithm set constraint sum account chang bound count constraint avg met preserv knapsack problem structur refresh tupl avg accomplish solv knapsack complex refresh section queri set knapsack capac assign weight tupl shown column label figur knapsack optim tupl refresh tupl tr bound avg aggreg queri join comput bound answer aggreg queri join express multipl tabl select sql join express select predic compar column tabl method determin membership tupl appli join predic select predic classif express sql queri optim system standard join possibl incorpor special index discuss section tupl refresh significantli difficult presenc join tupl contribut refresh subset base tupl subset shrink answer bound depend affect classif combin effect aggreg column base tupl potenti contribut multipl aggreg refresh base tupl aggreg tupl affect aggreg tupl interact make problem complex consid heurist algorithm tupl refresh join queri investig exact complex problem hope find approxim algorithm tunabl approxim algorithm statu futur work implement bound aggreg function refresh algorithm present implement cooper discuss section underwai addit test algorithm realist plan studi choic bound width appendix affect refresh plan investig altern method bound function paper repres initi work replic numer avenu futur work divid futur direct addit function tupl refresh improv perform avail issu addit function expand class aggreg queri algorithm aggreg median preliminari result extend result handl group bound enabl count uniqu queri handl nest aggreg function requir understand bound result aggreg affect outer aggreg aggreg queri idea expand encompass type relat queri constraint run find lowest latenc path network node node constraint requir answer return latenc select distanc answer user express rel absolut constraint rel constraint express constant denot absolut constraint actual answer difficulti advanc base bound deriv pass cach data find conserv absolut constraint algorithm redesign algorithm perform rel bound probabilist system defin paper improv perform provid bound offer absolut discuss section approach improv perform give probabilist interest direct combin provid bound answer probabilist appli idea replic data object resid sourc hierarchi data cach refresh occur cach cach sourc level cascad effect current scenario web cach system inktomi traffic server resid web server web browser extend data visual techniqu advantag investig wai extend data visual system displai imag base bound data draw fuzzi region uncertainti visual set model continu queri constraint formul visual domain tupl refresh adapt refresh algorithm refresh batch account multipl refresh cost sum individu cost adapt refresh algorithm account case refresh tupl reduc cost refresh tupl adapt develop refresh algorithm queri involv join express refresh tupl group join reduc subsequ cost join iter refresh algorithm set tupl advanc adequ actual exact refresh tupl iter constraint met addit develop altern suit interest investig context iter method prefer batch method present paper iter method give bound aggreg queri behavior user present bound answer gradual refin time goal shrink answer bound fast improv perform delai propag insert delet data cach investig wai discrep number tupl comput bound answer queri account bound discrep sourc longer forc send refresh time object insert delet investig special bound function suitabl updat pattern properti bound function shape suggest paper base assumpt inform updat pattern wai amort refresh cost refresh piggyback refresh sourc extra refresh request extra refresh consist valu refresh close edg bound amount refresh piggyback perform depend benefit versu ad overhead benefici perform send unnecessari refresh system load low futur process investig queri process issu bound valu design evalu scheme index bound function time discuss section plan weigh advantag function bound versu potenti index improv bound constant plan studi wai cach data object store pair bound function compress cach store valu data object sourc transmit valu tupl refresh refresh monitor sourc track bound function remot cach data object compress issu address affect techniqu present refresh algorithm independ bound function queri process issu consist issu handl refresh delai network refresh delai time master exce cach bound time cach refresh cach bound short period time avoid problem close edg bound evalu concurr control solut refresh permit occur refresh comput queri evalu answer reflect inconsist data fail satisfi constraint solut implement concurr control permit refresh occur queri read data current queri start acknowledg hector rajeev discuss joe hellerstein anonym comment initi draft marti discuss network monitor refer effici data share inform retriev system intern confer extend page march grahn queri set world acm sigmod intern confer manag page san barbara demarc techniqu maintain linear arithmet constraint distribut system intern confer extend page march goodman concurr control recoveri system queri constraint object acm sigmod intern confer manag page san libkin exact approxim aggreg constraint queri languag acm symposium principl page constraint objectori system acm sigmod intern confer manag page june rivest introduct algorithm mit widom comput median uncertainti acm symposium theori shasha danger replic solut acm sigmod intern confer manag page june garei johnson comput guid theori freeman concret foundat comput scienc gibbon summari statist improv approxim queri answer acm sigmod intern confer manag page june haa interact data analysi control ieee august scalabl trigger process intern confer data page march hellerstein haa onlin aggreg acm sigmod intern confer manag page wolfson diverg cach architectur intern confer parallel distribut inform page septemb kim fast approxim algorithm knapsack sum subset problem journal octob inktomi inktomi traffic aggreg approxim queri process page april revesz constraint queri languag acm symposium principl page april kuper aggreg constraint workshop principl practic constraint rhode april jr semant issu connect incomplet inform acm transact septemb comput base manag inform system embodi answer accuraci user paramet phd uc berkelei comput scienc freytag integr heterogen inform system intern confer larg data septemb stonebrak acm sigmod intern confer manag page june poosala fast approxim queri answer precomput statist ieee intern confer data page march rundenstein bic aggreg possibilist fifteenth intern confer larg data page august silberschatz relat data model eighteenth intern confer larg data page august jiang move object issu solut tenth intern confer scientif statist page juli good bound function return issu allud section briefli discuss good bound function select make problem good bound function separ shape bound determin static call width paramet bound data sourc assum monoton function time data object lower bound upper bound width paramet chosen algorithm problem face task select function algorithm width notic repres pair bound function ad benefit encod current width paramet transmit sourc cach refresh time cach comput time refresh delai sourc transmit refresh time clock synchron neglig threshold term function absenc inform model chang data object random walk dimens model natur common set updat tend small increment decrement current random walk model constant amount time step probabl distribut binomi distribut varianc inequ upper bound probabl distanc start fix probabl unit start point function time bound probabl proport vari tight bound approxim shape function shape bound function bound function form curv figur illustr function vari width sketch dynam algorithm bound width paramet attempt minim number refresh avoid refresh updat master bound wide make exce bound avoid refresh constraint bound narrow chanc type refresh chanc obviou bound width minim total probabl refresh requir factor affect choic bound variat data valu make result function neg deriv note randomli vari move step function rapidli grow continu move function dramat requir user queri difficult adapt algorithm adjust condit chang strategi start time refresh occur signal bound send bound time refresh occur signal bound strategi find middl ground wide bound exce exceedingli narrow bound refresh constantli fluctuat futur plan refin detail suggest techniqu perform experi determin eventu balanc conflict requir queri updat plan bound function case updat pattern conform random walk model proof correct refresh recal section refresh algorithm tr tupl min constraint show choic tr correct show tupl tr solut suffici constraint show tupl tr solut tr refresh tupl refresh tupl result upper bound answer ha la ha min min definit tr min ha la constraint hold tupl tr solut constraint hold show tr suffici constraint min tr note tr min min tupl tr min ha refresh tupl tr bound width case occur tupl tr refresh valu equal comput exact minimum valu greater la ha ha la comput max comput max select predic max aggreg function symmetr min ha max refresh algorithm tr tupl max comput max select predic max aggreg function symmetr min ha max max algorithm tr tupl max classifi tupl select predic algorithm section requir classifi tupl belong predic user assum arbitrari boolean express involv binari comparison defin transform predic transform yield express find tupl possibl satisfi predic base bound valu transform yield express find tupl satisfi predic base bound valu appli find tupl find tupl tupl predic evalu tupl tabl express term attribut valu endpoint min attribut valu rang handl express assum valu case constant attribut min max min max figur set translat boolean express translat rule appli queri select predic obtain note disjunct conjunct implic translat classifi tupl translat classifi tupl case misclassifi tupl extrem unusu involv special case correl note misclassif affect optim correct algorithm illustr rule figur deriv express term rang endpoint predic min max max min column label figur show result classif tupl data figur turn techniqu part gener mathemat framework introduc evalu predic data object set valu infinit set express min max min max min max max min min max min max min max min max min max max min min max min max min max max min figur translat rang comparison express point rang relationship translat notat paper notat select predic influenc evalu section inform select predic classif tupl slight refin made select predic column aggreg special tupl restrict actual select addit bound bound addit restrict predic contribut result advantag addit bound tupl shrunk input result comput refresh algorithm aggreg latenc predic latenc modifi lower bound comput tight bound avg select predic comput tight bound avg select predic section proce sl kl sum cardin low valu repres tupl sort order element sl kl add sl kl advanc continu process sl kl sh kh repres tupl sort order element sh kh add sh kh advanc continu process sh kh ha sl kl sh kh queri section averag latenc link classifi tupl shown figur initi sl kl tupl sort order refin appli select predic restrict aggreg column predic ta tb restrict column ta sl kl set sl sl kl kl sl kl set sl sl kl kl note sl kl stop sl kl comput sh kh proce similarli yield sh kh result give bound avg sl kl sh kh algorithm comput tight bound avg run time log refresh tupl avg select predic refer section note make bound avg section constraint ha la satisfi ha desir find refresh algorithm constraint satisfi exact bound ha sl kl sh kh appendix find algorithm set tupl tr refresh tupl tr comput sum count make bound sum count formul function base set tupl tr refresh valu constraint met relationship sum count rh show case case rh case rh case sum rh case rh sum sum sum sum count case rh rh impli case rh rh impli case rh rh sum bound impli goal express function elimin valu substitut conserv estim valu conserv estim valu obtain comput sum count current cach bound sum sum count count refresh bound shrink sum sum count examin inequ relat sum count substitut sum sum count make side strictli upper bound substitut result sum sum sum sum count count final upper bound function sum sum count comput substitut function constraint recal constraint requir substitut formul constraint term rewrit constraint term count sum sum sum sum count formul constraint place origin constraint algorithm free valu satisfi reformul constraint reduc task refresh tupl avg task refresh tupl sum reformul constraint refresh tupl constraint constraint count sum sum sum sum count function count sum sum sum sum count refresh tupl sum function recal algorithm refresh tupl sum map knapsack knapsack capac make knapsack capac function modif make knapsack capac function constant make knapsack capac function fake recal equal number tupl refresh remain refresh recal set item knapsack correspond tr set tupl refresh equal number tupl knapsack knapsack initi knapsack capac count time tupl ad function line sum sum sum sum count capac knapsack amount time tupl ad knapsack observ knapsack capac item ad weight item simul shrink knapsack time add quantiti weight tupl algorithm algorithm map knapsack count tupl sum sum sum sum count valu sum sum count found comput sum count current cach bound sum sum count count run time domin run time section revisit section queri averag latenc link classifi tupl shown figur comput sum sum count count valu assign weight tupl comput weight tupl ad sum sum sum sum count column label figur show weight knapsack problem count knapsack optim tupl refresh tupl tr bound avg 
Agents/3067 1 specif simul system caselp maurizio martelli mascardi zini disi di genova itali abstract nowadai softwar applic great complex exist compon properli integr distribut involv entiti heterogen make adopt technolog paper present prototyp environ system caselp prototyp method set tool languag support prototyp realiz system specif architectur descript languag adopt describ prototyp term agent provid requir servic commun link agent specif execut languag easili defin reactiv proactiv agent linear logic languag defin sophist agent system oper implement primit defin extend target languag simul tool integr caselp visual prototyp execut collect statist logic prototyp introduct nowadai applic strongli complex heterogen distribut entiti cooper coordin exchang share knowledg paradigm emerg technolog face problem model kind applic suitabl model entiti commun monitor environ react event occur initi situat human agent societi agent call system account distribut involv agent integr heterogen softwar fundament aspect success todai softwar system time evid engin approach build applic due inher complex applic prototyp approach classic model greater flexibl paper present caselp applic specif environ base logic logic program base prototyp environ ma develop caselp start year ago appli research ma rapid prototyp mean logic program applic model period time understand extend framework featur progress ad cope emerg necess caselp profit develop prototyp area applic relat transport logist problem develop collabor fs solv problem freight train schedul railwai la milano involv plan good realiz intern compani servic autom applic concern retriev medic inform distribut case caselp successfulli adopt engin process final combin constraint logic program techniqu appli manag distribut transact paper structur section introduc caselp agent section describ method realiz prototyp section focu languag tool stage prototyp develop section present tool languag model phase prototyp section describ instrument implement phase section relat execut phase section conclud paper outlin relat futur work agent caselp caselp agent repres perspect descript point caselp agent interpret set servic kind identifi agent facilit manag logic agent provid control coordin ma compon reason capabl interfac agent provid interfac extern softwar modul agent ma facilit agent suppli agent servic manag agent creat delet agent applic architectur refer intern architectur agent architectur facilit manag agent fulli integr caselp specif simul system caselp data structur form agent intern compon flow control master compon activ interpret depend type extern modul interfac agent link interpret approach softwar integr ma adopt caselp extern modul written program languag servic defin function agent requir accomplish servic typic depend domain applic prototyp provid servic export extern user ma servic import ma export import servic repres interfac ma extern world function ma requir extern human mass export import servic subset provid requir servic set caselp prototyp support export servic provid order interact human user prototyp main set export import servic empti abstract architectur comput point view caselp agent form behavior engin engin compon state includ data structur repres current situat comput behavior repres knowledg agent accomplish provid servic agent behavior rang form simpl rule complex plan engin encod task control govern flow control agent activ interact agent outlin featur caselp framework level languag tool environ method guid user prototyp inform descript applic develop work prototyp step user adopt suitabl languag tool provid caselp prototyp method realiz softwar prototyp perform step figur graphic method static architectur descript prototyp develop decid static structur ma step broken determin class agent applic determin kind agent belong eventu interpret set servic determin set instanc defin interconnect instanc link appropri request provid servic phase defin agent form commun channel notic architectur refer test start end static definit interact definit specif verif test implement figur develop method ture librari architectur defin descript commun agent provid request servic specif agent servic requir step sequenc exchang messag start properli captur step model ma includ set relationship specif system agent stage specif languag model prototyp level model specif interact agent abstract architectur account interact model step specif architectur chosen step model interact intern compon specif agent behavior verif test system step verifi perform preliminari valid abstract specif defin step specif simul system caselp valid specif languag previou step execut verif proof mechan specif languag specif logic execut proof procedur prototyp specif initi follow evolut ma detail verifi comput carri comput start configur lead final state satisfi properti implement prototyp step transform abstract specif applic final prototyp extend concret languag prototyp implement interfac extern softwar data provid interpret messag pass detail abstract previou implement execut obtain prototyp prototyp execut valid prototyp client desiderata execut user develop inform event occur check prototyp behav expect encompass desir featur error discov step impli choic made previou step distinguish phase method adopt caselp step form model phase prototyp step repres implement phase step repres execut phase section focu languag tool caselp face phase list prototyp specif section describ languag tool prototyp model simpl architectur descript languag step method adopt custom architectur descript languag ma describ class agent involv ma interconnect instanc architectur descript inform interpret servic agent class templat definit agent class multipl instanc class defin servic provid agent instanc link servic requir instanc provid agent servic link singl agent instanc commun structur agent set specif link tool model step method describ model ma agent commun languag subset kqml model defin sequenc messag perform content messag start middl imagin agent request servic agent requir accessori servic agent order repli agent relat sc defin eventu start fm messag sc denot start messag handl receiv agent denot eventu start messag handl receiv agent start receiv agent hhf linear logic languag hhf execut languag model concurr resourc sensit system base gener specif logic languag forum hhf logic combin featur extens logic program languag goal implic notion formula resourc basi linear logic hhf collect atom linear disjunct correspond head linear implic kind consum resourc order appli resolut step hhf applic extend type gm solv order trigger specif simul system caselp refer step caselp develop hhf level model ma hhf specif prove check properti commun agent instanc deadlock architectur hhf specif includ detail agent data rule model architectur engin behavior specif architectur depend rule defin hhf descript observ evolut agent backward detect violat requir specif hhf framework ma defin model ma agent heterogen architectur note caselp aim provid set specif languag compil prototyp implement languag ma develop pick suitabl familiar present hhf mandatori choic specif system adopt specif languag altern hhf part futur work librari architectur specif specif agent architectur difficult phase develop method develop caselp furnish librari architectur hhf pick desir model librari specif proactiv specif procedur reason system fulli languag program behavior proactiv agent behavior reactiv agent express mean rule proactiv agent behav rule order execut agent trigger percept implement messag condit satisfi agent state action updat agent state send messag agent system behavior agent obtain mix proactiv reactiv rule syntax sketch behavior behavior rule rule rule messag msg check condit check condit msg perform condit true goal true goal goal goal goal action assert retract agent initi state set atom ground formula reactiv rule fire messag agent condit action execut condit form distinct condit agent express state order execut sequenc action auxiliari set call auxiliari procedur call auxiliari condit satisfi sequenc action execut auxiliari procedur defin agent procedur prototyp proactiv step develop method altern hhf limit behavior specif level notic syntax extend behavior specif complex agent perform prototyp implement step method deal implement version prototyp target languag implement extend prolog languag extend prolog develop agent defin standard prolog extend commun capabl safe state updat predic agent updat state ffl assert assert fact agent state implement safe predic assert fact remov state ffl retract remov fact agent state unifi fact fact predic agent inspect read messag ffl sync block call agent messag unifi messag enter ffl receiv msg search call agent mailbox messag unifi msg input messag msg output unifi msg output unifi atom messag caselp initi develop success implement sicstu prolog assum languag prototyp automat compil develop sicstu prolog specif simul system caselp ffl receiv fetch messag call agent messag unifi list messag predic agent send messag ffl send messag call agent agent receiv send predic implement safe updat predic assert state retract state predic call execut updat agent implement work prototyp implement work proper code suppli agent system code implement comput featur initi behavior respect specif agent system step method interfac code defin interpret predic abil automat translat agent specif code implement avoid burden prototyp implement develop scratch research direct recent result complet automat translat result obtain engin pr architectur hhf approach translat hhf rule code hhf pr engin assum gener form gm formula atom order translat guard introduc auxiliari predic pc defin retract retract assert retract state assert state predic state updat subsect execut retract retract consum atom formula proof test guard condit current final execut assert assert add inform state appli transform hhf set obtain set partit group main activ perform pr agent plan plan execut action execut engin agent defin predic cycl cycl defin pr engin result cycl plan plan action execut activ defin belong activ plan trigger pc pc plan trigger group issu consid compil due specif easili translat form manipul engin work compil input program gener code form reactiv proactiv reactiv rule reactiv rule proactiv rule pm predic cycl agent form cycl receiv msg output messag select reactiv select proactiv predic cycl defin simpl engin agent architectur inspect messag pick fire reactiv rule select agent behavior set action return select proactiv rule set proactiv action return set action chosen execut sophist strategi rule select action choic complet engin integr caselp compil engin proactiv architectur complet specif simul system caselp prototyp execut agent implement piec compil target languag load make code run simul compil agent primit implement file file agent differ small portion state behavior output primit file sicstu prolog modul modul system develop load initi start simul follow execut mean graphic issu command sicstu prolog shell caselp graphic simul interfac caselp visual document event happen agent level ma execut develop code agent automat instrument load instrument add probe agent event relat state exchang messag record collect visual window initi agent set simul length visual granular provid view relat instrument agent shown end simul complet trace instrument event visual instrument complet independ execut influenc futur chang execut support figur show trace execut prototyp train traffic manag shell command simul execut develop prefer type command set primit execut simul load agent agent initi initi agent global min time set global simul paramet iter number iter simul express unit min time unit file exchang messag flag set messag visual monitor simul run net failur min max coupl interact failur rate simul stop visual event occur period interrupt occur occur decid develop figur trace execut minimum maximum latenc commun channel failur rate set lost due commun problem min latenc max latenc minimum maximum express delai commun channel coupl commun net paramet channel assum delai complet reliabl start simul start execut prototyp built load respect paramet simul schedul activ turn agent call cycl predic simul execut primit statist collect inform set start end time length interv simul period number assign monitor messag file statist collect simul number messag receiv interv sum weight total number exchang maximum averag number messag maximum averag weight weight standard deviat provid inform prove evalu commun channel fundament import applic commun bandwidth limit specif simul system caselp relat futur work tool method constitut caselp repres approach softwar prototyp potenti approach demonstr adopt prototyp technolog develop applic area choic logic program base prototyp environ proven discuss modular express lp extrem describ clear fashion complex behavior agent prove model ma capabl agent support integr extern softwar environ ma specif implement base logic languag logic program adopt kowalski sadri develop architectur unifi ration reactiv interest object form architectur repres logic formal agent task control perform execut reaction proof procedur cycl wagner approach similar previou defin vivid agent vivid ration ad reactiv agent behavior repres action reaction rule exploit logic program realiz applic base ma found schroeder al present formal implement diagnost agent base extend logic program logic formal defin agent congolog concurr program languag base logic theori restrict languag event action model intern mental state agent behavior futur work aim improv aspect environ evalu applic distribut simul simul point caselp central global time agent system improv effici caselp chang simul distribut cope distribut caselp agent equip addit data structur agent simul activ process local simul engin ad agent implement life consist inspect commun channel intern event list receiv integr specif languag legaci softwar set specif languag tool anim execut provid set languag tool interfac mean interfac agent augment extens face project base rapid prototyp environ good global inform integr research experi base logic includ common joint project lead develop gener open framework rapid prototyp engin softwar develop applic hot area wider field applic ma technolog web extend caselp ontolog order semant integr data typic situat explor collect data web develop web model research direct begin acknowledg part research realiz zini visit depart comput scienc softwar engin melbourn refer di sistemi logica linear master disi di zini logic program synerg combin applic semant logic program paradigm perspect zini system develop softwar engin proc san januari springer verlag lnc zini agent base prototyp freight train traffic manag proc workshop event septemb zini combin logic agent rapid prototyp engin distribut applic proc septemb delzanno logic program linear logic phd di dipartimento di genesereth ketchpel softwar agent commun georgeff lanski reactiv reason plan proc de hj levesqu concurr program languag base situat languag implement februari specif simul system caselp kowalski sadri unifi agent architectur combin ration reactiv proc intern workshop logic san kramer distribut softwar engin proc intern confer softwar page ieee comput societi press zini appli logic program specif complex applic proc june mascardi constraint technolog distribut transact manag proc june icsc academ press finin evalu kqml agent commun languag intellig agent ii lnai miller specif logic theoret comput ndumu nwana research develop challeng system iee proc softwar de caselp tool di sistemi master disi di di master disi di rao bdi agent speak logic comput languag van de veld agent break page lnai de almeida pereira delib reactiv diagnosi agent base logic program intellig agent page lnai wagner artifici agent logic program proc post confer workshop logic program page juli wooldridg jen intellig theori practic knowledg engin zini sterl design ontolog agent proc septemb 
ML/750 6 autom state abstract option algorithm ander andrew barto depart comput scienc massachusett ma abstract learn complex task significantli facilit defin hierarchi subtask agent learn tempor abstract solv assign accomplish task studi hierarch learn framework option argu full advantag hierarch perform state scale larger state abstract autom adapt mccallum algorithm automat build state featur illustr result algorithm simpl hierarch task result suggest autom state abstract attract approach make hierarch learn system effect introduct research field reinforc learn recent attent tempor abstract action term tempor abstract describ action variabl amount time motiv tempor abstract action exploit hierarch structur problem hierarch structur natur incorpor prior knowledg learn system tempor abstract action polici learn task learn hierarchi significantli reduc number situat learn agent discrimin framework option extend theori reinforc learn includ tempor abstract action accur execut option polici depend state featur learn agent featur relev differ option option hierarch learn perform state abstract irrelev featur specif option state abstract hierarch learn system save memori develop compact state learn gener induc abstract dietterich introduc state abstract hierarchi tempor abstract action approach requir system develop defin set relev state featur action prior learn complex problem increasingli difficult state remedi problem autom process construct state appli mccallum algorithm individu option achiev state abstract algorithm automat build start make distinct observ vector specif depend prior learn section give descript algorithm section introduc modif make algorithm suitabl learn hierarch system describ setup experi section present result section section conclud discuss futur work algorithm algorithm retain histori transit instanc observ time step previou receiv transit previou instanc base compon assign uniqu leaf tree distinct leaf determin path pair algorithm action estim futur discount reward execut util leaf denot max algorithm model consist estim transit probabl jl expect reward comput transit instanc model perform sweep iter execut modifi valu pair jl reinforc learn algorithm updat action priorit sweep algorithm period add distinct tree form temporari call fring perform statist test ad distinct predict power distinct base perceptu observ previou histori indic back current histori dimens examin leaf tree extend fix construct permut distinct path leaf instanc leaf distribut leav ad fring distinct statist compar distribut futur discount reward leaf node polici action fring node polici action distribut futur discount reward node polici action argmax estim futur discount reward individu instanc jl ks test output statist differ distribut node algorithm retain distinct leaf sum ks statist differ fring node larger sum ks differ exce threshold tree extend leaf distinct tree action valu previou leaf node pass leaf node restrict number distinct agent make time limit depth length histori algorithm retain depend tree size size state set algorithm potenti scale larg task previou algorithm learn compact state satisfactori polici complex drive task version algorithm suitabl continu state space develop successfulli robot soccer adapt algorithm option turn issu adapt algorithm option hierarch learn architectur finit markov process state set option consist set state option polici choic termin condit probabl option termin state reach primit action gener option termin time step easi defin hierarchi option polici option select option local reward function option facilit learn option polici make algorithm suitabl perform state abstract simultan defin state polici separ assign algorithm perform state abstract separ option modifi polici option level hierarchi oper time transit instanc form make scheme add notion tempor abstract definit transit transit instanc option form observ vector time step option execut option termin time durat discount sum reward receiv execut previou instanc option level hierarchi execut experi sequenc transit instanc algorithm work option histori instanc base distinct instanc algorithm develop task option termin execut histori made finit segment separ execut option transit figur taxi task instanc record execut independ instanc record previou execut updat segment algorithm appli hierarch learn option learn option oper part state space learn option behavior gener execut option process call learn action valu option updat base action execut option updat occur action execut probabl execut base distinct option transit instanc record execut option ad instanc record execut option histori option associ instanc vector approach requir addit memori multipl copi instanc scheme introduc vector reward fr instanc discount sum local reward option experi test version algorithm taxi task grid taxi assign task deliv passeng locat chosen random set site taxi agent observ vector locat current passeng destin action taxi cardin direct passeng passeng random pickup site reward provid taxi deliv passeng illeg action move aid taxi agent introduc denot set observ vector locat polici agent learn introduc local reward ident global reward provid agent except reach applic algorithm taxi histori option maximum length instanc length oldest instanc histori discard expand tree consid instanc histori set depth threshold distinct present case algorithm lower threshold agent make distinct difficult case accumul evid statist differ accept distinct algorithm back reconsid distinct import reduc number incorrect distinct due statist evid implement compar distribut futur discount reward leav instanc taxi task fulli set histori index algorithm system pick random action probabl perform softmax tune softmax temperatur good balanc explor improv perform avoid part ensur action execut regularli design set experi examin effici learn randomli select option randomli select posit taxi reach ignor issu deliv passeng begin learn assign singl node option set algorithm regular learn option share instanc set polici option taxi task learn parallel polici task option action reward provid task sum global reward local reward option execut passeng taxi posit select randomli passeng randomli select pickup site result result learn experi shown figur graph learn regular learn averag independ run tune set learn run give maximum perform interv time option save evalu separ evalu consist fix repeatedli navig target time randomli reposit taxi time target repeat ad reward conclud learn converg faster regular learn achiev higher level perform faster converg due fact histori option fill quickli learn higher perform achiev amount evid larger target option reach execut reach time execut option set perform learn durat time step regular figur comparison regular learn time step figur show result node repres distinct drawn leaf node shown squar omit denot distinct execut option order letter denot distinct observ note make distinct lower part grid branch algorithm made suboptim distinct distinct smaller number leav suffici repres optim polici figur total leaf node number leaf node vari averag leaf node make actual number state smaller compar result dietterich state compar distinct state flat distinct state polici requir result improv memori requir store histori account memori save due state abstract larger task significantli outweigh memori requir shown algorithm option hierarch learn system result suggest autom state abstract perform algorithm attract approach make hierarch learn system effect testb import step autom state abstract hierarchi incorpor learn method learn agent extract inform train data result show learn significantli improv perform learn agent perform state abstract main motiv develop hierarch version algorithm autom state definit transit instanc enabl histori structur learn solv problem partial observ domain futur work examin perform state abstract algorithm realist task plan develop version task figur polici algorithm back tree reconsid distinct potenti improv perform algorithm correct node incorrect distinct made acknowledg author tom dietterich provid code taxi andrew mccallum valuabl correspond ted perkin read provid comment paper work fund nation scienc foundat grant recommend express materi author necessarili reflect view nation scienc foundat refer hierarch reinforc learn maxq function decomposit artifici intellig research state abstract maxq hierarch reinforc learn muller advanc neural inform process system pp cambridg mit press emerg hierarch control learn relationship reinforc environ mea matar anim animat cambridg mit press reinforc learn select percept hidden state phd comput scienc rochest reinforc learn hierarchi machin solla advanc neural inform process system pp cambridg mit press model tempor abstract plan solla advanc neural inform process system pp cambridg mit press reinforc learn hierarchi abstract model proc nation conf artifici pp menlo aaai press learn tempor abstract action proc intl conf machin pp morgan kaufman mdp framework tempor abstract reinforc learn artifici intellig gener adversari reinforc learn aaai fall symposium model direct autonom system 
DB/1461 3 semant xml dtd text data mine techniqu abstract unstructur document textual archiv public patient record annual report share inher structur order facilit search archiv enabl inform integr text collect relat data inher structur made explicit detail infer semant structur xml document type definit archiv subsequ transform text xml document success method achiev object main contribut paper method deriv structur xml dtd order extend deriv flat dtd framework deriv unstructur xml dtd compon support larg number document xml tag preliminari dtd priori assum mandatori fix order xml tag automat tag archiv deriv dtd implic tag error introduc notion probabilist xml dtd compon assign probabl semant structur correct method establish probabilist xml dtd base discov associ resp frequent sequenc xml tag keyword semant dtd knowledg data cluster introduct organ cope huge amount text document tan point inform store unstructur textual document captur interest action knowledg textual major challeng data mine commun creat semant markup form provid explicit knowledg text archiv facilit search enabl inform integr work author fund german research societi grant sp winkler £ spiliopoul leipzig graduat school manag depart germani relat data sourc user manual creat metadata due effort cost involv text mine techniqu requir automat creat semant markup tag document present kdd approach pursu research project german acronym stand integr legaci system structur document mean data mine goal semant tag textual content facilit identif integr text relat data aim deriv structur xml dtd serv document collect enabl provis queri servic textual data text collect vocabulari syntax frequent share structur framework semant tag domainspecif text introduc appli workbench text archiv result collect semant tag xml document extract unstructur xml dtd ultim aim integr result xml document relat data sourc deriv preliminari dtd transform structur dtd reflect order option tag xml tag deriv data mine techniqu iter cluster explain section crisp due tag error critic fact introduc notion probabilist dtd describ order xml tag statist properti tag structur dtd basi futur inform integr effort involv xml archiv gener workbench introduc algorithm infer probabilist dtd util associ rule discoveri algorithm sequenc mine techniqu rest paper organ system der von und da von und eur gesellschaft mit der ist und ist zum die der gesellschaft im tabl xml document annot commerci regist entri section briefli discuss relat work section overview framework semant tag domainspecif text collect section introduc notion probabilist dtd textual archiv develop method deriv conclud give direct futur research section relat work moonei combin method kdd inform extract perform text mine task appli standard kdd techniqu collect structur record featur text feldman al text mine term level linguist tag word author repres document set term construct taxonomi term result dataset input kdd algorithm associ rule discoveri framework adopt idea repres text term concept goal semant tag structur text unit sentenc document global dtd entir content loh al suggest extract concept individu word subsequ kdd effort document level similarli author suggest exploit exist vocabulari thesauri concept extract describ workbench acquir domain knowledg text similar approach combin method field research unifi framework approach share research thread object extract semant concept text concept extract serv element xml dtd discov concept peculiar singl text unit suffici perfectli reflect content order deriv discov group text unit share semant concept concentr significantli differ averag text respect word frequenc statist collect process standard text mine softwar integr relev domain knowledg prerequisit success knowledg discoveri research activ aim transform text semant annot xml becker al introduc search engin ess support queri process text deriv process xml text abstract abstract summari text separ origin text provid semant text intact process visual linguist applic domain dtd content document infer content order transform exist content xml method infer dtd tag document input method tag plain text document deriv dtd closer approach work keyword rule convert legaci data xml document approach reli establish rule base drive kdd methodolog reduc human effort data topic relat research commun lot effort recent put method infer repres structur similar document approach deriv schema set document simultan solv problem text document semant tag infer appropri structur xml dtd describ relat archiv awar scientif commerci approach emploi probabilist document type definit introduc paper describ text archiv integr text relat data sourc framework notion semant tag refer activ annot text xml tag addit attribut classifi entir document tag singl aim semant tag text unit sentenc paragraph tabl illustr concept semant sentenc german commerci regist entri text unit semant sentenc made explicit xml tag partli addit attribut describ extract entiti person amount xml document depict tabl creat appli framework collect textual commerci regist entri text unit collect includ entri relat foundat compani district german citi compani oblig law submit inform busi affair local commerci regist commerci regist import sourc inform daili busi textual content search queri moment semant textual archiv basi inform integr creation servic relat inform brokerag xml queri languag emploi submit queri semant tag xml archiv framework pursu object archiv text text document semant tag preliminari flat xml dtd deriv archiv semant tag dem process design knowledg discoveri textual process constitut phase order build cluster semant similar text tag document xml result deriv xml dtd describ archiv kdt process introduc result final set cluster label serv xml tag dtd element huge amount document convert xml document product phase framework text unit document cluster built text unit cluster subsequ tag cluster label concentr semant tag similar text document origin common domain approach semant tag kind archiv public announc court administr quarterli annual report textual patient record health care applic product servic descript publish electron marketplac text document entiti text unit cluster accept cluster xml document type definit person date corpor place currenc uml schema thesauru entiti descript nlp preprocess creation text unit extract replac entiti select featur unit map text unit featur vector set paramet execut algorithm evalu cluster qualiti cluster inspect unaccept cluster refin cluster semant label accept cluster xml tag text unit creation dtd xml document xml document figur iter interact kdt process remaind briefli introduc phase framework iter interact kdt process depict figur process term cluster algorithm invok repeatedli notion iter cluster fact cluster algorithm perform multipl pass data converg process knowledg engin consult cluster evalu final cluster end iter initi text document domain knowledg constitut input kdt thesauru taxonomi term preliminari uml schema domain descript specif entiti person compani uml schema reflect semant entiti relationship initi conceiv applic expert schema serv refer dtd deriv discov semant owner tabl preliminari unstructur xml dtd commerci regist entri final dtd schema similarli convent kdd process start preprocess set level granular determin size text unit workbench perform basic nlp preprocess normal word stem remov stop establish drastic reduc featur space select limit set term concept text unit thesauru uml schema text unit descriptor chosen knowledg engin reflect import concept applic domain text unit map boolean vector featur space entiti interest extract text unit separ modul workbench case creat small thesauru select relev descriptor point descriptor pattern discoveri text unit vector initi archiv cluster base similar content object discov homogen text unit cluster cluster perform multipl iter iter output set workbench partit qualiti criteria cluster text unit vector cardin larg text unit homogen semant small number text unit descriptor member cluster subsequ remov dataset remain text unit vector input data cluster algorithm iter cluster similar threshold cluster progress specif content process base concept execut cluster algorithm workbench case emploi demograph cluster function includ ibm intellig miner data maxim criterion workbench discov altogeth cluster approx text unit phase consist label cluster assign label cluster label determin knowledg engin workbench perform rank candid cluster label expert default cluster label deriv featur space dimens text unit prevail cluster cluster label correspond xml tag subsequ annot cluster member origin document tag valid xml tag xml tag enhanc attribut reflect extract entiti valu tabl excerpt unstructur xml dtd automat deriv xml tag case studi describ semant structur result xml collect entiti serv addit attribut xml tag fulli evalu workbench establish probabilist dtd output workbench set semant xml tag xml tag describ content archiv document reflect content archiv abstract essenti tag dtd semant annot deriv data mine crisp essenti valid tag express quantit term estim properli order tag deriv semant xml combin probabilist dtd deriv order tag comput statist properti tag insid document type definit reader recal semant annot label cluster discov workbench underli cluster mechan produc nonoverlap cluster impli text unit belong effect annot label cluster deriv workbench nest extens workbench hierarch cluster algorithm establish subclust nest label plan futur work object dtd establish method specif order identif correl mutual tag adorn tag correl tag statist properti properti form basi reliabl queri determin expect recal queri result introduc statist properti dtd tag associ describ methodolog comput statist model complet statist inform pertin tag hypergraph structur introduc mechan deriv probabilist dtd graph statist properti semant xml tag statist properti dtd tag depict tabl paragraph column properti column reflect properti peculiar set tag cluster label tag group tag column mechan appli deriv properti tag accuraci workbench deriv semant xml tag label cluster cluster constitut model convent statist term data model subject misclassif error identifi type error type text unit assign wrong cluster label reflect content text unit error type text unit assign cluster label reflect content text unit envisag error type relev term accuraci model probabl cluster label reflect content cluster member accuraci affect dtd peculiar individu tag incorpor statist adorn individu tag order evalu qualiti approach absenc drew random sampl text unit domain specialist verifi annot text unit respect error type error type type occur text unit tag text unit correctli process percentag error type ii text unit indic text unit cluster semant belong error rate entir dataset interv result tag dtd cluster label deriv statist approach term observ option se applic domain expert provid suggest tag observ mandatori suggest hold true text unit inform misclassifi inform simpli absent document expect movi movi unknown due natur movi properti offer indic tag consid potenti mandatori defin ratio document tag total number document archiv associ miner identifi item occur interest tag affect tag term tag ü tag ý confid defin associ rule ratio tag ý ü document ý similarli associ correl tag ü set tag ý high support ü popul statist call lift improv defin allevi ratio ü ý ü popul ratio ì ü statist properti tag order tag account order tag essenti term tag ü sequenc adjac tag ý number document sequenc ý number document ý ý ýò definit differ convent statist sequenc mine concentr adjac disallow occurr arbitrari tag convent sequenc mine satisfi requir web usag miner design distinguish adjac event frequenc tag frequenc group set sequenc term properti radiu comput method accuraci model workbench tag simpl statist set tag associ rule discoveri set tag associ rule discoveri sequenc tag sequenc mine set sequenc tag ratio number document group tag total number document set properti assum set valu perturb set member show model statist inform pertin individu tag group set relationship seamless model statist xml tag valu statist properti depict tabl made part workbench remain comput data mine algorithm exploit valu establish probabilist model algorithm build dtd process model introduc gener graph statist inform group tag tag relationship depict structur establish dtd xmlschema rich statist adorn discuss algorithm dtd establish repres tag associ direct graph node individu sequenc adjac tag set tag node adorn statist properti pertin resp tag group edg repres relationship form ý convent ü sourc node group node lh target similarli edg adorn statist associ repres semant tag graph node set semant xml tag deriv î ¢ set graph node conform ì ì tag graph consist fact xml tag deriv kdd method group distinguish group perform step model tag group order tabl statist deriv xml tag list annot list flag group depict list order list irrelev uniqu permut group tag map requir list lexicograph order è î set list element î ü è î form ú ú list element î list repres group ü repres uniqu group ú î î tag annot lexicograph group distinct group element group group element group group list violat lexicograph order list element è î defin î è î ¢ î group threshold threshold input mine usual kdd set low threshold affect size graph execut time algorithm build dtd tag relationship graph edg set node constitut graph î î indic node singleton tag group tag statist edg eman element î point element î tag group tag set edg subset î ¢ î statist properti refer sourc group node target target sequenc adjac locat confid valid statist associ confid lift target set locat confid statist properti assum null graph properti compon group tag relationship adorn statist valu tag discov workbench present graph group tag present depend threshold group support conceiv minimalist approach high frequent group approach tag combin occur document present opt minimalist graph connect gener case group tag frequent frequent relationship tag rare combin tag contrari approach ensur combin tag document depict graph case document singl tag occur document size graph depend threshold group support confid lift valu approach deliv upper limit comput ñ number label output workbench ò ñ largest number distinct tag document è ò perturb result equal number ò ò group edg tag particip maximum result ñ ñ ¢ graph node ñ ¢ edg upper limit graph size threshold valu statist properti essenti obtain manag graph cutoff impli inform loss observ graph approach refer structur introduc algorithm deriv probabilist dtd construct part graph dtd deriv graph version captur relationship semant tag found workbench similarli process schema establish convent design decid relationship entiti worth captur refer statist presum dtd reflect relationship present document rare graph relationship set sequenc adorn partial statist present algorithm deriv dtd construct part graph tag dtd adorn probabilist refer tag locat insid dtd algorithm heurist deriv concentr pair tag frequent prefer maxim sequenc tag reader recal comput statist relationship tag requir activ data mine softwar algorithm back miner return desir statist backward construct dtd sequenc algorithm observ dtd set altern sequenc build sequenc start tag algorithm build maxim mean tag sequenc tag document support sequenc backward algorithm start arbitrari tag identifi tag tag sequenc expand anymor mark algorithm shift sequenc arbitrari tag tag process predecessor prepend sequenc iter predecessor subsequ built found predecessor altern incomplet sequenc produc duplic sequenc built algorithm process iter predecessor tag found invok sequenc miner return frequent sequenc adjac tag algorithm frequent pair ü lead confid ù tag predecessor hold è ù ratio document predecessor divid total number document maximum ù determin rest element sequenc sequenc mark maxim criterion mention larger ü predecessor sequenc case maximum margin larger valu tag show maximum locat confid differ locat confid tag small tag accept result altern subsequ identifi maxim algorithm consid longer frequent sequenc return sequenc end subsequ built frequent sequenc found subsequ check maxim impli comput ratio document start number document compar ü tag support comparison perform guidelin altern tag document start maxim document start adher maxim sequenc statist maxim final algorithm filter sequenc maxim assign probabl valu tag insid maxim sequenc multipli accuraci model output workbench locat confid tag respect subsequ lead backward versu forward sequenc construct method gener altern sequenc dtd tag prune frequent sequenc adjac tag produc sequenc miner method construct advantag incorpor sequenc core miner categori perform forward sequenc mine kernel modifi expand sequenc successor tag dtd tree altern algorithm observ dtd tree altern subsequ adorn tag support respect subsequ lead insid number document start subsequ tag similarli algorithm tag predecessor observ dtd tree impli common root gener document archiv start tag assum dummi children tag document tree node refer tag children refer tag context predecessor dtd tree altern resembl dataguid statist adorn differ algorithm consid sequenc tag document frequent observ complet sequenc miner return arbitrari subsequ tag method realiz preprocessor modul web usag miner modul coerc sequenc event common prefix place tree call tree input navig pattern discoveri process perform core sequenc tag document observ sequenc effect preprocessor build dtd archiv tree altern tag sequenc figur depict tree relat case studi note xml document depict tabl partli dtd excerpt knowledg hidden electron media organ encapsul document acquir knowledg impli effect queri document combin inform piec textual asset function confin queri text search engin scan individu asset return rank result present methodolog enabl queri process join text sourc structur deriv xml dtd text archiv mean data mine techniqu semant text unit core approach deriv xml tag undertaken workbench part studi main emphasi combin tag reflect semant text unit archiv singl dtd reflect semant archiv shown dtd probabilist proxim archiv content deriv set statist properti reflect qualiti tag insid dtd relationship tag statist properti tag relationship form basi combin complet dtd xml xmlschema graph structur depict statist serv basi oper mechan deriv dtd emploi mine algorithm set heurist rule test methodolog archiv document region commerci regist deriv set tag workbench implement mechan deriv dtd futur work includ implement mechan dtd deriv establish framework comparison deriv dtd term express accuraci ultim goal work establish queri mechan text archiv intend coupl dtd deriv method queri mechan data dtd deriv probabilist impli design model evalu qualiti queri result acknowledg german research societi fund project mbh provid data project collabor stefan conrad discuss ibm intellig miner data kindli provid ibm term ibm scholar program figur dtd tree altern tag sequenc refer suciu data relat semistructur data xml morgan kaufman san agraw srikant mine sequenti pattern proc int conf data mar hugh navig pattern discoveri internet data page neumann construct linguist search index internet search engin intern confer applic natur languag inform june berri linoff data mine sale custom support john wilei buneman semistructur data sixteenth acm symposium principl page staab manual semant text annot tool journal section semant zamir text mine term level european symposium principl data mine knowledg page septemb mine web navig path fragment goldman widom enabl queri formul optim semistructur page aug winkler framework convert text xml document data mine techniqu ieee intern confer data san spiliopoul semant tag text document intern workshop inform fusion page editor workshop web mine challeng aug acm schema find regular semistructur data principl data mine knowledg european volum lectur note artifici page septemb heidelberg oliveira knowledg discoveri text extract web acm sigkdd im masand editor advanc web usag mine user lnai springer juli workbench acquisit ontolog knowledg natur languag seventh confer european chapter comput page march moonei inform extract aid discoveri predict rule text sixth intern confer knowledg discoveri data mine workshop text page august manolopoulo effect predict data mine approach workshop mine log data custom san august motwani infer structur data sigmod schmid probabilist tag tree intern confer method languag page septemb transit exist infer document structur der xml xml meet page spiliopoul labori data mine web mine int journal comp sci special issu mar spiliopoul tool web util analysi extend version proc edbt workshop lnc page springer tan text state art challeng workshop knowledg advanc page april wang liu discov structur associ semistructur data ieee transact knowledg data 
HCI/1018 4 futur multimedia user interfac blair steven feiner columbia matter power eleg technolog underli multimedia user interfac ultim determin system provid overview emerg area research plai major role futur multimedia virtual environ ubiquit comput discuss describ relationship current multimedia system user interfac provid access virtual environ involv interact address spatial modal research goal provid video ubiquit comput requir multimedia network infrastructur support extrem interact larg number mobil wireless user arriv understand potenti introduc terminolog explain basic system architectur lead discuss kei current research issu futur direct introduct glanc recent confer interact reveal multimedia activ area research commun work current desktop comput environ workstat video displai keyboard audio goal examin companion set research area exist hardwar technolog peopl interact futur comput physic place comput occupi live function network technolog displai extrem larg graphic remain option handwrit speech recognit matur except wireless network connect comput stationari comput comput peopl gener justifi led mental singl supplement portabl comput capabl price current commod justifi situat chang major paradigm eventu redefin interact comput virtual environ ubiquit comput virtual environ world creat coupl interact devic displai power multimedia comput ubiquit comput term coin mark weiser describ futur surround everydai life multitud comput bare examin work field explor head review terminolog provid histor overview field reli larg larg hardwar provid introduct import hardwar issu descript kei approach system architectur current research build background provid section lai set current research issu direct futur work attempt wai virtual environ ubiquit comput complement creat excit form multimedia comput power approach make multimedia system multimedia system futur multimedia user interfac terminolog term refer concept discuss articl begin prefer term virtual world virtual environ commonli virtual realiti term virtual realiti technolog deliv media mean describ user interfac rang physic environ present displai ordinari graphic displai convent game virtual world virtual environ refer gener comput system interact devic attempt provid spatial tactil devic displai stereo displai visual hand haptic describ artifici realiti term coin krueger intend describ virtual environ refer virtual environ requir user wear special hardwar experi control virtual environ krueger built system base project displai track vincent augment realiti refer displai overlai graphic user view real world current system hmd al feiner al reason displai user head track enabl system render accur graphic regist properli real world refer precursor virtual environ hmd coupl pair camera remot locat user head remot camera move give strong system claim provid partial feel system user perform task typic manipul master control remot slave effector mimic master control movement slave effector signal back master forc feedback telerobot system control robot remot ubiquit comput refer environ larg number rang hundr unit seamlessli integr connect wireless network comput commonplac awar tool accomplish task weiser draw analog electr motor major small cheap awar motor dozen electr motor oper electr ignit ubiquit comput peopl comput track move comput adapt transpar prefer person notion comput comput awar user prefer respond predict unlik comput central focu simpli assist ubiquit comput comput stai background peopl concentr task ubiquit plenti term past small comput ambiti person digit assist begun market current system offer function popular imag devic head exemplifi appl hypothet knowledg navig devic intim detail inform intellig assist owner manag daili life enabl softwar technolog futur pda agent agent small program perform specif activ behalf comput user comput program part popular videotap present knowledg navig earlier scienc fiction inspir agent depict human persona impli broad rang skill person assist butler comput augment environ term pierr wendi rich gold al refer technolog add comput power real world virtual ubiquit multimedia tional broad rang physic machin ubiquit comput overlaid graphic project displai displai virtual ubiquit multimedia propon ubiquit comput weiser vision notion virtual attempt make world insid note applic benefit strongli virtual game virtual walkthrough inaccess argu averag user weiser term refer virtual environ todai opaqu hmd glove shut user rich real world version technolog virtual environ limit virtual environ gener plai major role futur ubiquit comput extend vision ubiquit comput includ descend technolog current augment realiti research hmd point peopl wear bulki hardwar daili basi treat todai hardwar improv predict peopl carri comput weigh comput todai bulki hmd replac attract displai displai potenti soni earli centuri virtual environ ubiquit comput inher multimedia paradigm virtual environ involv auditori tactil compon combin creat convinc virtual world world involv materi comput record materi case augment surround real world ubiquit comput natur integr tradit digit graphic video larg number displai potenti intens multimedia significantli current technolog entir displai devot video connect imag left long period time audio import displai identifi sound call attent import inform minim issu address convent multimedia network oper system vital success virtual environ ubiquit comput histor overview section briefli highlight import mileston histori virtual environ ubiquit comput histori virtual introduct ubiquit interest exampl virtual environ comput morton develop machin call demo user virtual ride york address visual auditori haptic invent stereoscop hmd auditori stimul earli flight simul creat virtual environ aid comput live gener shoot model board tv camera ivan sutherland gave talk ifip confer notion room object convinc real object develop virtual environ hmd sutherland system present vector graphic small crt worn side user head optic path fold mirror make augment realiti system inspir sutherland talk ultim fred brook began work haptic displai present forc combin test prototyp model applic futur multimedia user interfac earli michael built prototyp forc feedback devic posit small knob vertic shaft motor control forc exert knob user hold knob move cube explor simul surfac vari dynam hard start krueger built number system carefulli avoid technolog user bulki wearabl displai opt displai track approach trade physic comfort appeal research nasa al develop kei idea modern virtual environ includ color stereo hmd spatial sound foster mid research creat dataglov popular gain media attent dozen virtual environ develop consult research form root ubiquit comput lie histori person begin pioneer work alan kai colleagu xerox parc power comput intend children adult interest pda grew research pda dai serv intellig agent perform varieti search bui airlin ticket prospect world fill anthropomorph assist ubiquit comput research began xerox parc late explor altern view place futur comput occupi everydai life hardwar technolog section provid overview hardwar current virtual environ ubiquit comput spatial haptic displai interact devic offer includ detail reader understand capabl user interfac spatial tracker varieti spatial track system exist measur report inform joint angl sensor provid posit orient inform tracker technolog popular electromagnet ultrason electromagnet produc polhemu stationari transmitt mutual orthogon cube rang inch foot side coil sequenc electr signal produc magnet field smaller cube inch coil transmitt magnet enabl system determin receiv posit orient depend receiv track distanc figur popular spatial track technolog electromagnet technolog flock courtesi ultrason fly courtesi spatial tracker twenti feet transmitt electromagnet tracker requir direct line sight tracker sensit metal object magnet field ultrason stationari transmitt small sourc arrang emit seri ultrason receiv small microphon base delai transmit system determin distanc speaker rel posit microphon orient receiv recov ultrason tracker affect presenc metal magnet sensit requir direct line sight transmitt receiv accuraci electromagnet ultrason tracker leav measur posit small fraction inch orient small fraction technolog requir cabl receiv electron box optic track system video camera track object approach optic track refer system posit camera world object track accur track control condit system requir object track visual aid attach led mark system attach camera object track watch world move posit orient camera object determin exist system work gener environ north carolina chapel hill tracker ceil tile grid led al technolog track posit orient explor includ magnet creat wearabl system requir extern transmitt receiv track accomplish worldwid global posit system gp receiv receiv signal posit satellit maintain signal determin absolut posit gp receiv antenna earth accuraci report posit inform ten meter accuraci obtain posit addit receiv transmitt posit broadcast correct base differ comput posit techniqu call differenti gp result accuraci meter accur kind gp system compar phase gp signal stationari receiv phase signal receiv move achiev accuraci real time figur north carolina chapel hill optic track system ceil user instrument grid led system comput user head posit base set led visibl camera mount head ronald north carolina chapel futur multimedia user interfac track system base wearabl activ badg al activ small infrar transmitt period transmit uniqu identifi owner badg transmiss detect short rang receiv room hallwai build monitor make room section hallwai occupi person wear badg ir fairli direct requir clear user desir privaci simpli put badg prevent receiv pick transmit signal instrument dataglov al dexter hand provid inform finger joint angl dataglov accomplish set ten joint hand finger cabl run loop back damag pass assign joint tini led end cabl measur amount light pass cabl finger bent joint measur light provid indirect measur joint angl posit orient part glove measur electromagnet sensor technolog build glove appli track user part system finger track prototyp acceleromet finger develop system wearabl solid surfac serv chord keyboard approach make track hand bodi requir user wear special apparatu gener comput solut eventu process user silhouett recogn hand bodi pose real time vincent wellner technolog develop decad infrar light illumin user ey camera track dark reflect infrar rest ey system measur ey orient requir user wear camera posit front stationari camera altern approach monitor electr muscl system discrimin larg area user entir visual field system contend involuntari ey movement user fact ey motion part normal ey point result unfortun consequ graphic displai input graphic displai creat virtual environ categor surround viewer basic kind provid presenc fill larg amount user field displai surround displai hmd mount user displai view special design user focu displai comfort figur displai stereo displai design view ey hmd conjunct tracker materi view depend posit orient user head make build virtual environ surround user move surround displai system surround head accomplish build room possibl floor displai approach make surfac video screen al deer shown figur provid sensit head system stereoscop displai system head posit tracker current stereo displai technolog suit applic reli liquid crystal opac control impress voltag left ey imag altern displai screen doubl standard frame rate control synchron left len transpar left ey imag len transpar ey imag displai distinct opaqu displai opaqu displai block user view real world present user visual environ displai overlai graphic displai input graphic user view real world opaqu displai stationari current displai optic combin overlai graphic merg video imag world graphic opaqu displai optic combin mean overlaid graphic user real world full field view visual system provid opac control level individu pixel reduc user visual experi real world resolut field view opaqu hmd introduc addit lag video circuitri hmd camera locat posit user imag world offset displai rang convent crt displai combin exploit technolog surround system term virtual world desktop virtual world system term deriv cubic volum insid crt describ group work ubiquit weiser mention basic class tini held palm point ubiquit comput virtual environ present polar displai virtual environ plai import role ubiquit comput describ lcd fit movabl window inform space classifi ubiquit comput virtual environ devic techniqu appli natur displai creat richer inform space ubiquit comput virtual environ emploi wide rang interact devic vari displai class ubiquit comput devic discuss pen input figur displai virtual figur cave user wear liquid crystal stereo glass perceiv effect imag project directli floor mirror fold optic path video projector minim amount space cave carolina futur multimedia user interfac suitabl input devic displai tradit keyboard devic enter larg amount data virtual environ research explor kind input devic techniqu includ simpl hand gestur recognit instrument digit imag design shown figur figur parctab displai ubiquit comput environ low resolut pen button infrar link wireless commun xerox pad size displai data system figur xerox liveboard displai xerox figur interact devic virtual environ flat front face left cricket microphon ultrason track system arrai small circular bump handl cricket haptic feedback digit imag audio displai input audio displai input virtual environ ubiquit comput comput graphic displai input research area explor sound virtual work concentr audio output express properti virtual difficult speech input replac gestur ubiquit emphasi integr portabl comput wireless telephon servic creat person commun current multimedia sound output treat companion graphic present flat displai stereo audio wed screen graphic creat compel visual sound listen surround audio environ similar imag object space depend distanc orient rel viewer sound object function distanc orient rel listen ear case left ear hear sound left ey object differ sound listen make audio clutter parti listen pick attend ongo listen mono main cue hear sound direct differ sound amplitud arriv time left ear strong direct cue account abil hear sound head sound sourc equidist ear reach time equal sound travel circl front asymmetri asymmetr head shape sound spectra depend direct sound sourc rel head filter call transfer function insert small microphon subject ear measur waveform real sound sourc locat posit mathemat model develop sound spatial system foster digit signal process technolog process sound sourc real time appli simplifi version index desir locat sourc rel ear result stereo pair signal present listen ear sound sourc eman rel locat simpl system model locat sound sophist includ addit effect absorpt reflect room surfac haptic displai haptic displai address user touch addit sensat forc torqu interact convent input devic keyboard touch forc import difficult touch type membran keyboard kei offer tactil feedback tradit keyboard kind passiv haptic feedback direct result physic keyboard mechan construct control comput haptic displai activ comput control build mention section number research system provid exampl power haptic displai motor driven held user biologist experi molecular forc simul molecul jr al explor surfac sampl scan remot electron inspect modifi featur atom level control microscop ii al joystick produc feel patch textur materi al end simpl tactil displai form arrai small pin dramat tactil feedback user cricket architectur softwar architectur virtual environ ubiquit comput evolv set goal virtual environ system develop support singl hardwar platform small number tightli coupl intent creat applic futur multimedia user interfac system origin model tradit interact program evolv solv problem virtual environ exist system ignor issu applic larger requir realist virtual world work develop infrastructur support virtual environ network oper system level work ubiquit began vision environ futur prototyp system built design address infrastructur issu process model virtual environ earli virtual environ applic simpl head hand handl sequenti problem approach frame rate tightli coupl time iter main event sum time requir read input process step virtual environ gener displai frame moder complic virtual time process step simul nontrivi singl timestep scientif visual minut virtual environ displai respond track user ten time simul determin displai user head posit determin obviou solut problem distribut system multipl decoupl comput step simul loop al result displai updat fast head posit inform displai continu fast step current virtual environ distribut process tightli coupl process master creat initi process system easili reconfigur fault toler difficult construct master applic configur control devic problem kind encount share draw messag traffic demand virtual environ typic volum larger number user interact devic data commun support distribut simul allevi dv al dive structur applic larg set process process independ commun messag protocol problem exist system scale gracefulli build virtual distribut comput load significantli larger number machin current system design handl exist system distribut tend central comput virtual environ simul detect commun result interest parti fundament problem central algorithm scale huge number primarili bandwidth commun media quickli overwhelm larger number machin inform result comput system address problem scale arpa simul network project al develop hundr thousand simul train militari person collabor tactic recent al similar run graphic hardwar distribut interact simul protocol understood consist dozen network packet type heterogen simul commun uniqu featur protocol larg number particip veloc inform posit messag object particip simul vehicl typic travel fairli predict posit object calcul local reliev send entiti continu error real calcul posit exce posit inform broadcast approach success hand head motion human move predict wave system system develop goal deal requir larg scale distribut virtual environ commun media vari bandwidth gener fashion infrastructur ubiqu infrastructur ubiqu main compon ubiquit comput mobil comput softwar infrastructur ubiquit wireless network tie network requir discuss section current rate technolog comput displai requir met decad leav softwar infrastructur develop wireless infrastructur suitabl test displai scheme network rout activ explor al ioannidi maguir weiser gener oper system issu deal heterogen network discuss al problem includ deal widespread effect intermitt connect wireless handl intermitt connect filesystem mobil comput transpar local cach allevi problem cach applic determin file oper result applic unpredict user oper investig deal structur cach filesystem inform inform user oper file issu inform applic environ chang custom interfac appropri applic told nearest explicitli printer approach problem modifi semant environ introduc notion dynam environ variabl automat inform interest applic valu chang al applic simpli configur print system inform applic local printer variabl ubiquit comput environ capabl devic comput creat offic environ user intellig control surround make automat adjust light heat system base prefer room occup al al addit system save amount energi scale turn equip room displai gener model popular imag virtual realiti fulli graphic current technolog provid peopl feel ultim goal displai gener research strive environ potenti provid visual complet complet unnecessari inappropri technolog achiev fulli virtual environ futur discuss displai gener model virtual fulli render augment realiti imag process fulli render scene straightforward common approach creat virtual environ render fulli frame user viewpoint model virtual world user head paramet synthet camera user ey scene render tradit comput graphic techniqu scene render stereoscop scene render point view ey small displai close user fix respect user head ey constantli move respect world desktop displai user head fix respect constantli move rel user head ey conceptu major differ user standpoint graphic differ displai small differ paramet suppli synthet camera render graphic kei featur environ virtual world clever approach reduc number complex object render frame textur reduc number polygon user creat system obviou disadvantag approach extrem futur multimedia user interfac power comput creat realist environ real time augment realiti altern fulli graphic render select object combin user real world unlik fulli entiti exist real world render highlight suffici draw wirefram box circl shown figur virtual world commun wealth inform annot real world creat hundr line polygon al user cut rich real mundan object pen paper peopl comput imag process render entir scene viewpoint render wide imag scene singl posit imag process techniqu produc imag render viewpoint user pose work object distant user user posit roughli arbitrari direct potenti import approach broadcast video game nearbi object ad render separ simpli overlai distant background closer rest scene share live video imageri roller ride overlaid imag interior roller gener user inform gener share virtual experi broadcast group peopl research issu direct number issu address virtual environ ubiquit comput current approach obviou improv track processor commun network substanti improv virtual environ ubiquit comput hope gain widespread accept time technolog technic pursu time figur augment realiti addit world render mainten assist al user shown pull printer paper trai current desir posit paper trai highlight line style arrow drawn show user direct trai move columbia hardwar requir interact techniqu develop paradigm autom assist applic hypermedia agent support collabor privaci area ripe explor hardwar requir import hardwar issu mobil comput speed comput wireless network connect work network portabl comput power wireless network mobil virtual environ applic increasingli limit track displai technolog speed comput network primari limit system tracker lag ubiquit import hardwar factor physic size comput bandwidth wireless network processor continu faster requir power hold true tradit displai technolog sort requir tab pad bandwidth speed wire network advanc futur ubiquit comput virtual avoid discuss specif requir ubiquit comput virtual environ differ mainstream applic tracker virtual environ ubiquit comput complet track system virtual environ requir extrem ubiquit comput long wireless system system deliv high tether system versu low wireless system total rang track system virtual environ compar smallest unit accuraci typic ubiquit import mind paradigm good advantag system ubiquit comput environ custom base occupi room higher environ custom base displai problem current track discuss section categor accuraci rel import attribut depend specif head tracker hmd locat tracker virtual requir differ depend system augment realiti low latenc user head track azuma point rang accuraci requir complet virtual environ stringent requir virtual suffic approxim posit orient user visual overrid conflict signal rang tracker user navig virtual world physic move desktop virtual environ system requir rang user close stationari track user head feet displai room displai suffici augment realiti requir extrem high accuraci low latenc ensur virtual imag regist properli real world ubiquit potenti applic requir track scale magnet ultrason system scale transmitt rang limit requir locat matter feet object environ produc requir filtrat research long highli accur track optic track project mention section track system scale exhibit uniform high accuraci low latenc entir rang bishop major drawback current system size room requir camera head ceil restrict head problem insurmount major bulk hmd due size weight video bigger miniatur camera gener visual track algorithm futur multimedia user interfac requir room instrument remov addit benefit make track system singl transmitt receiv conceiv gener track algorithm determin posit dynam object conjunct inerti tracker gp wireless locat track provid track system unlimit rang optic track techniqu uncommon find comput displai attach video camera camera track person posit rel captur video inform determin posit user head gener imag displai person point view track inform section track system latenc improv critic import virtual environ al azuma bishop measur track object move data structur displai affect user experi manifest augment delai result graphic real world current system head motion predict extrapol filter reduc appar posit orient lag al unc optic hmd estim addit acceleromet predict orient posit head displai displai requir ubiquit comput devic power higher resolut interest requir simultan multipl pen address liveboard al liveboard color displai system multipl function meter displai surfac group interact natur user reach displai surfac capabl displai displai virtual environ present set current solut current hmd readi widespread obviou requir hmd resolut field view approach fidel human visual system ey field view approxim vertic ey provid horizont field view approxim stereo hmd pixel arc minut visual angl requir ey displai resolut human visual constant finest detail resolv field view receiv field pixel wide resolut render graphic high resolut entir visual field visual fall rapidli object move rang vision displai wide field view low resolut simultan displai user gaze equal effect approach electron figur imag ax overlaid corner object world improperli regist imag properli regist imag ronald north carolina chapel processor bundl transfer imag light valv projector hmd ey present imag field view ey higher resolut field view direct regard user ey displai present great challeng render softwar ensur visual artifact introduc object move region hmd made lighter displai eventu construct individu pixel transpar opaqu subtl problem depth field displai pixel singl imag stereo pair displai addit forc visual system deal conflict depth visual clutter focu time problem notic augment realiti virtual object focu real object colloc focu displai construct individu pixel depth program control processor video plai larger role ubiquit comput virtual manipul video stream video present produc show vision comput interfac point video window displai remot particip rotat perspect give particip effect real time broadcast video effect accomplish digit video stream multimedia comput graphic processor textur map surfac overhead graphic pipelin creat textur data structur raw imag preclud modif textur earli hardwar network success ubiquit comput hing develop wireless network exce capac plan commerci system decad commerci system aim connect fairli wide rang huge number ubiquit comput devic suggest significantli higher bandwidth network support design provid high volum unit area weiser suggest build wireless demand requir aggreg bandwidth easier achiev network provid high bandwidth unit area cellular system cell size radio frequenc hundr time attempt build long rang wireless network figur rotat imag remot particip video particip rel posit bruce futur multimedia user interfac devic displai video tradit bandwidth requir order magnitud higher video preval virtual happen displai hardwar begin support section virtual environ lower bandwidth stricter latenc requir data millisecond latenc perceiv user lag system movement combin virtual environ ubiquit comput virtual environ ubiquit comput put forward compet complementari obviou area overlap involv virtual realiti peopl run applic ubiquit comput environ graphic assum track technolog matur state ubiquit comput devic user track reason advantag inform displai graphic weiser augment realiti displai ubiquit comput live room window overlai user view path neighbor friend morn weiser user head regist graphic imposs graphic gener viewpoint viewer assum displai base current displai technolog viewer graphic viewpoint chang track user move problem solv individu hmd advantag ubiquit comput environ materi present person devic requir displai simpli notifi infrastructur virtual inform user present ubiquit comput devic displai individu custom number simultan user current ubiquit comput research care peopl disabl person track disabl ubiquit comput control environ al al user turn affect advantag augment limit visual displai activ audio displai environ annoi graphic displai person spatial audio avoid pollut sonic environ user investig accommod larg number comput user interfac physic visual virtual world feiner hybrid user combin real virtual displai common user interfac displai person comput virtual inform surround present hmd prototyp hybrid user interfac window flat panel serv primari work space window move freeli flat panel inform space provid context work extend account larg number varieti devic popul ubiquit comput feel concept hybrid user interfac kei effect integr ubiquit comput augment realiti approach address problem discuss support collabor colloc user augment realiti section place video video plai increasingli import role multimedia interfac advantag integr video camera highlight section potenti fix track peopl front displai addit provid inform gener posit inform determin portion video imag transmit video applic ensur transmit imag view user achiev camera mount move enabl remain point user technolog bederson spheric point motor accur posit camera mount rotat speed hundr render camera enabl transfer stereo imag camera posit appropri remot viewer reason stereo pair imag displai construct dynam model visibl part user head transmit recipi system displai remot user desir al varieti approach develop captur depth map environ al make mix graphic view unfamiliar environ visibl relationship properli maintain technolog tradit video integr fulli ubiquit comput virtual abil map video stream discuss section facilit place video stream virtual environ arbitrarili posit render mention section major current system strive render virtual world approach commun inform technic illustr found architectur illustr structur typic line draw view vari convei clearer understand object build realist render scene discuss autom approach gener imag hidden edg render line style depend object hide approach render import augment realiti convei inform minimum augment graphic feiner develop method automat gener view al al describ system produc imag inspir architectur render virtual realism import scene amount render time devot object percept hidden user graphic system long varieti facil build graphic object adapt transform posit size object cull scene render lower level base size project render requir virtual environ research suggest cull imag degrad techniqu base physiolog human percept pausch integr exist user interfac paradigm research tradit displai remain relev ubiquit comput virtual environ kind inform effect displai techniqu displai surround ubiquit comput environ similar wai kind displai discuss ubiquit comput virtual environ benefit incorpor major paradigm examin direct common techniqu virtual environ shift attent gener issu surround wimp studi class interfac discuss techniqu adapt accomplish environ manag direct manipul direct manipul refer interact techniqu object interest visibl user manipul directli increment action complex command languag direct manipul tout featur virtual environ natur learn reach grab conveni effect direct manipul turn critic tediou command line interfac replac problem compound poor qualiti current tracker glove complet lack haptic feedback virtual environ perfect hand user grab object manipul impli object arm length object user easili grab fast move futur multimedia user interfac small present similar problem direct manipul requir user hold hand physic support extend period time long date back light complic fact screen rest hand solut problem give user select object point problem farther object harder select virtual user chang distanc manipul pointer explor techniqu work object extrem larg quick difficult manipul directli wimp exist wimp techniqu applic adapt fit newer rang handheld manner familiar devic requir modifi techniqu direct wimp techniqu virtual environ attempt typic success simplest interact devic user physic devic virtual world virtual practic virtual keyboard poor track lack haptic feedback button pad successfulli emul posit set larg virtual button plane physic desk provid haptic feedback passiv physic prop instrument tracker provid user facil control virtual object al adapt wimp techniqu virtual panel system provid mr toolkit al mr panel essenti arbitrarili orient rectangl hierarch text manipul accomplish base intersect rai cast user hand surfac panel due problem current tracker compound lack haptic user experi problem panel analog exist techniqu found virtual world virtual environ advantag user tradit input keyboard mice monocular system pointer devic straightforward treat view plane larg displai cursor move manipul scene object manipul line sight cursor al devic stereo system straightforward task pointer tradit techniqu work straightforward line sight pass user ey system input straightforward approach devic cursor move view plane link eman fix point user ey intersect cursor ubiquit comput voic mice keyboard larg bodi research exist input techniqu tablet comput run oper microsoft window pen comput shafer common pda newton handwrit recognit gener major complaint comput print write slower form input keyboard research investig wai speed handwritten input goldberg richardson develop input techniqu specif design alphabet call significantli recogn input higher handwrit print richardson altern charact correspond gestur determin placement set pie menu commerci simplifi alphabet improv print pda appl newton techniqu faster accur print tradit requir addit train understood differ input techniqu requir size displai manner radic point interfac interfac reli thing reserv portion displai widget adapt small environ manag user point specif area displai top bottom edg adapt larg larg small displai manner chang current graphic user interfac base displai size sheet share displai applic remain displai creat effect assumpt reach import user window system provid window manag user manag larger set window fit displai exist window system inappropri environ microcomput window system microsoft window appl macintosh desktop requir program run local make unsuit distribut environ popular network window provid suffici gener layer abstract applic window system window move displai mention move displai size capabl attempt made modifi rectifi problem al satisfactori window move virtual ident modifi transpar window mobil difficult achiev window nelson provid layer abstract applic window address requir move window significantli displai develop suitabl window system provid capabl exist system challeng environ manag previou discuss problem encount attempt current window system ubiquit comput virtual environ prime motiv exist window invent desir support huge number exist futur applic al adapt window system issu address involv window manag screen real estat surplu real estat spread hundr devic user organ window small crowd organ order magnitud object scatter environ extend notion window manag environ manag simpl problem current window manag forc user posit window manual automat posit window nontrivi displai ubiquit comput augment potenti displai real environ manag freedom make reason choic object placement augment potenti overlaid object block displai person al environ manag user avoid problem move exist virtual object approach window manag interfac metaphor explor wong al displai set document stack pile paper set document typic convei exist group approxim content rel order pile metaphor advantag abil form recogn pattern complic data document pile mean owner virtu shape manual label learn disciplin commun gener pai attent happen disciplin increasingli import user interfac attempt user learn disciplin older matur continu comput power improv interact readili advantag experi learn anim motion squash futur multimedia user interfac techniqu ungar perceptu cognit psychologist studi peopl understand sensori architect landscap design structur real space effect integr hypermedia al stress import make hypertext capabl oper separ hypertext applic extend notion hypertext link make link arbitrari physic virtual object solomon al develop prototyp augment realiti system hypertextu link creat window physic object environ make automat find note made book attach remind peopl locat similar calendar program attach date attach hypertextu link follow room author author materi individu medium difficult creat qualiti multimedia document requir individu media skill make media work effect effort author multimedia document rare attribut sole defici process research direct address difficulti autom design goal explor mckeown al multimedia gener project system ai techniqu creat explanatori text graphic fulfil set input coordin materi gener media make gener text refer gener graphic al author made harder media includ full set possibl dynam virtual environ feiner al address aspect creat virtual environ automat design desktop virtual environ present visual multivari account criteria interact control amount time displai render vari resolut data calcul displai compon guid user equip mainten task creat graphic overlai user view physic world design account posit orient user head object inform present redesign user move annot object annot physic obscur object major challeng develop robust virtual environ design system handl broader rang media product includ tempor anim speech agent situat design multimedia present complet framework develop tool assist creation present inform agent assist collect inform monitor electron newswir articl roam comput network gather inform remot charg carri task part owner agent provid assist video edit process unannot footag recogn find shot includ particip locat al suggest previou full set output possibl dynam virtual ubiquit comput environ greatli complic multimedia present select output devic displai possibl wide vari task knowledg ubiquit comput displai agent automat select displai migrat inform displai agent assist organ inform singl augment realiti overlai larg displai captur journal person experi captur journal person experi bush suggest scientist wear small camera microphon select photograph thing interest enter note annot vocal work lab technolog make extend vision captur user experi tini pair camera microphon frame pair glass improv compress techniqu memori cost size eventu make idea feasibl myriad issu search lifetim record experi collabor virtual environ ubiquit comput addit issu address current research collabor cscw support collabor user displai interact devic mobil particip collabor pda ubiquit comput devic provid high resolut displai displai share inform augment realiti applic support privat view inform virtual world share presenc mean user virtual environ classic metaphor work user view share inform privaci privaci issu ubiquit comput virtual environ track volum data collect weiser discuss implic wear activ potenti interest person long spend room build activ badg track user granular tracker potenti user time possibl employ inform pleasant implementor pai special attent ensur privaci maintain weiser suggest combin technolog social solut employ video camera comput activ monitor watch gener social legal climat discourag technic feasibl measur technic solut ensur privaci ubiquit comput environ succe hamper util current attitud person attitud adjust appropri global activ monitor individu obviou automat forward phone call interest applic exist automat diari gener record user area dai al make easier fill time rememb involv check check talk back innoc intent inform secur system place prevent approach uniform global avail provid inform make posit inform physic talk hallwai back rememb necessarili find meet possibl peopl pool inform reconstruct underscor weiser point technic solut privaci electron inform pool requir problem address social imagin inform servic pai peopl contact inform pool inform sell pai privaci requir interest interfac question easili part huge volum inform collect queri meet back state light possibl complic combin public privat futur multimedia user interfac privaci issu address technolog eavesdrop confidenti inform current interfac provid prevent displai read work public place protect easi read displai person sit front plane train augment realiti solut problem displai virtual imag wearer hmd possibl read shoulder elimin collabor environ previou select part person inform augment public displai confidenti ubiquit comput provid straightforward solut case document group discuss perform automat close inappropri enter room survei state art virtual environ ubiquit user interfac paradigm shape futur comput show complement provid capabl multimedia user suggest set issu multimedia research explor paradigm differ preval current multimedia part hardwar displai interact case virtual larg number comput integr ubiquit comput paradigm addit requir oper system network current multimedia applic virtual environ involv transmiss inform gener tracker small amount inform comparison digit latenc requir stricter multipl user interact effect ubiquit comput requir high bandwidth extrem short result extrem high aggreg bandwidth entir network consequ assumpt made multimedia support system reexamin rang current user interfac wimp current comput static network configur refer testb dynam virtual environ spatial sensor proc acm uist page graphic comput graphic acm siggraph annual confer page appl human interfac appl desktop interfac publish ma track requir augment realiti improv static dynam registr optic hmd comput graphic acm siggraph annual confer page merg virtual object real ultrasound imageri patient comput graphic acm siggraph annual confer page miniatur activ vision phd york ny technic report sound virtual realiti multimedia academ ma design interact multivari visual ieee comput graphic grasp realiti graphic serv scienc proc acm chi page refer project displai scientif visual comput graphic page atlant virtual world architectur proc ieee vrai page virtual realiti system proc ieee vrai page power ny cartoon user interfac comput graphic annual confer page toolkit develop distribut virtual environ proc ieee vrai page virtual design implement cave comput graphic acm siggraph annual confer page make virtual realiti experi virtual portal proc graphic interfac page knowledg navig video appl larg interact displai support group present remot collabor proc acm chi page xerox parc tech report de offic environ ubiquit comput offic comfort energi manag technic report xerox palo alto research center window window augment realiti proc acm uist page augment realiti autom gener coordin multimedia explan ieee satisfi visibl constraint dynam illustr visual hybrid user breed virtual bigger interfac physic smaller comput proc acm page virtual environ displai system proc acm workshop interact page situat inform space spatial awar palmtop comput interact visual medic data visual scientif page reprint ieee august pp wearabl interfac proc acm chi page futur multimedia user interfac seri posit track system promot literatur research system global posit system ieee stylu user interfac manipul text proc acm uist page stylu proc acm page program network portabl comput proc fourth workshop workstat oper system page system render proc acm chi page text input softwar appl newton palm lo ca virtual environ system proc comput graphic confer iri hypermedia servic apparatu individu unit state patent number file reprint comput page simul unit state patent number file august passiv interfac prop visual proc acm chi page design implement mobil internetwork architectur proc winter page scienc virtual realiti virtual environ ma enhanc treatment hidden line acm transact experi wireless environ proc usenix symposium mobil page militari network technolog appli theme park home entertain system comput microelectron person comput scientif make design architectur distribut virtual environ proc ieee vrai page artifici realiti ii publish ma make real world virtual user interfac issu mobil comput proc fourth workshop workstat oper system page realism virtual realiti environ proc acm uist page applic optic comput graphic comput refer manual technic digit system research refer centr research report pile metaphor support casual organ inform proc acm chi page gener multimedia explan proc aaai page microsoft window applic design guid microsoft wa feel issu forc displai proc acm symp interact page focu rang sensor proc intl conf comput page gener automat track technic xerox user interfac guidelin publish tactil commun sid reproduct human imag virtual space teleconferenc proc ieee vrai page issu interfac technic report xerox palo alto research center interact graphic displai architectur proc ieee vrai page intellig agent comput imag gener john york window system acm transact sc parctab mobil comput system proc fourth workshop workstat oper system page custom mobil applic proc usenix symposium mobil page decoupl simul virtual realiti mr toolkit acm transact inform optic design criteria binocular mount displai displai system spie volum direct step program languag ieee render frame influenc peopl comput graphic sun toolkit refer manual sun part ultim displai proc ifip page virtual realiti interfac scan tunnel microscop comput graphic page oper system issu pda proc fourth workshop workstat oper system page video prototyp case histori proc acm chi page futur multimedia user interfac properti specif achiev grace disconnect oper intermitt mobil comput environ technic report xerox palo alto research center interact dedic multimedia author system proc acm chi page alphabet proc acm chi page virtual world virtual applic explor academ press design illustr languag product influenc graphic gener proc european chapter assoc comput page activ badg locat system acm transact inform warp develop kit refer manual warp ca synthet visual environ hand gestur voic input proc acm chi page comput centuri scientif comput scienc issu ubiquit comput ubiquit comput ieee interact paper digitaldesk introduct realtim digit synthesi virtual acoust environ proc acm symp interact page illustr comput graphic acm siggraph annual confer page hand gestur interfac devic proc acm chi gi page construct virtual world proc acm symp interact page 
ML/1975 6 sensor dynam tactil inform peer eric rolf institut fur germani submit robot autonom system keyword tactil trajectori dynam tactil humanoid robot abstract present tactil grasp simpl robot gripper part consist arrai capacit coupl object mean brush fiber sensor element sensit threshold robust damag grasp yield type dynam tactil inform roughli type tactil sensor human skin complet sensor consist static forc yield total forc center forc distribut surround arrai dynam sensor element sensor mount gripper jaw humanoid robot equip electron bu interfac applic describ experi evalu qualiti grip sensor measur util guid robot arm desir trajectori neglig forc introduct robot compon describ paper exist work humanoid robot grasp humanoid problem camera locat distanc actual manipul place visual analysi inaccuraci reduc camera mount manipul setup call humanoid anymor start problem idea live visual inaccuraci fine tune grasp mean tactil end devic high sensit high spatial resolut import serv detect contact earli forc order minim probabl damag robot object environ vision suit actual contact point visibl analysi scene current state accur long robot equip suitabl set sensor execut task manipul everydai learn unknown analysi object shape surfac properti approxim friction grasp tactil sensor human skin place tabl human skin type tactil list properti tab import distinct sensor type static dynam sensor react constant pressur static react static pressur skin dynam roughli measur speed skin prefer react vibrat indent speed characterist due sheet appli mechan filter indent signal skin similar tactil receptor replac function replac hair cell dynam sensor special import activ check surfac properti rough surfac coin distinguish perfectli flat appli static pressur slightli surfac reveal inform immedi necess activ tactil object analysi explor object shape improv imperfect grip unexpect contact determin dynam begin contact tempor deriv high indent small static pressur sensor requir forc feedback grasp met combin dynam static element integr sensor modul similar patch skin technic tactil sensor section give review technic tactil sensor basic principl commonli capacit sensor consist plate capacitor forc sensor modifi plate distanc area shift rel posit made construct sensor dynam measur capac major reach order magnitud capac measur exampl sensor creat measur charg forc deform restrict dynam measur suit measur vibrat induct sensor measur chang chang geometri magnet coupl size sensor make technolog sensor tactil linear variabl transform engin wide sensor consist light transmiss medium photo detector pressur lead chang light intens transmiss regist detector sensor chang resist pressur attract measur resist requir electron metal semiconductor conduct call forc resistor wide reduc resist appli plastic foil sensor hardwar describ construct combin tactil sensor static sensor static part sensor foil base foil coat ink foil resist logarithm pressur produc interlink sell label hold patent techniqu measur amount posit forc appli place foil electrod foil electrod consist nest parallel contact finger contact finger electrod directli connect electrod connect linear measur resist foil resist yield forc inform appli voltag edg electrod measur voltag electrod inform posit forc direct order full coordin system put top static part similar sensor develop liu al measur forc rang dynam sensor element place figur order sensor arrai decid build capacit dynam sensor recept part sensor consist round capacitor membran prototyp microphon plate radiu rest distanc capacit pf microphon ic attach fix basic serv field transistor reduc capac fig recharg capacitor plate distanc chang time constant ms major problem solv coupl capacitor object element cover elast lead spatial filter contact inform requir high forc deform larg regist sensor disadvantag resolv mechan transduct techniqu brush fiber attach flexibl membran silicon contact object fiber transmit forc capacitor chang plate distanc choic membran time capacitor charg reflect approach speed actuat object number fiber attach capacitor paramet adapt desir applic prototyp fiber element due flexibl method suitabl transmiss static forc ideal vibrat detect contact small forc exert object place figur addit raw detect contact speed appli analogu filter circuit measur vibrat standard circuit discharg time constant ms call speed signal vibrat roughli correspond inform complet circuitri dynam element shown fig combin sensor place figur combin sensor consist static element surround dynam sensor element mount jaw gripper static element yield posit forc redund direct dynam element produc speed vibrat result total signal gripper electron gripper box attach gripper bodi fig place figur addit dynam sensor ad tip gripper jaw order detect contact tabl grasp setup adapt flexibl special layout sensor shown figur fig combin static sensor dynam sensor repeat potenti strongli miniatur provid sensit skin high spatial resolut process sensor signal analogu signal type sensor element digit process microcontrol siemen microcontrol convert analogu circuit construct multiplex signal gripper digit signal current setup channel leav potenti addit torqu sensor gripper joint bu made address sensor modul robot modul make arm patent applic sensor techniqu file full detail construct prototyp found experi sensor properti place figur present experi probe properti sensor experi suggest speed approach object estim raw sensor vibrat situat slide object detect filter vibrat signal sensor element rapidli brought contact object remov time signal shown fig speed signal reach maximum millisecond contact show restor origin voltag experi fig object approach element speed stop suddenli result vibrat reflect vibrat measur apparatu adapt earli phase slope signal strongli correl speed approach experi start object held stabl grasp gripper jaw close forc reduc object begin slide fig show time speed signal influenc turn detect vibrat signal robot place figur experi carri humanoid robot detail depict figur fig consist modular robot arm freedom kinemat similar human parallel jaw dual stereo camera head dof stereo basi cm camera pair field view color comput network pentium pc linux sun ultrasparc ii workstat solari pc reserv control robot perform imag acquisit color control camera imag process processor network achiev throughput low latenc manual guidanc robot arm approach behavior guid tactil implement method guid robot arm manual desir trajectori gripper orient open close implement dynam control except rotat axi gripper control gripper adjust fit object guidanc place figur speed signal sensor binar suitabl threshold trigger predefin reflex repres translat rotat movement vector rel gripper translat vector move contact posit target center gripper jaw result radial vector field direct vector point target posit sensor posit rotat vector turn gripper maxim number sensor attempt parallel orient gripper jaw hand surfac vector field shown fig open width gripper control contact hand case contact gripper jaw receiv small open case contact small close signal place figur control guid robot arm desir trajectori put hand ideal object parallel surfac jaw adjust movement compon sensor element add motion minim asymmetri contact hand hand movement configur space fig simpl static object parallel grasp strategi eventu converg situat symmetr contact distribut desir grasp posit gripper jaw close gripper turn method good grasp strategi minim requir object shape develop systemat learn optim vector field subject current research full detail dynam robot control found interpret sensor signal experi carri construct sensor evalu qualiti grip place figur fig show graphic user interfac written displai sensor output red circl central blue field show static center pressur denot center circl amount pressur radiu blue field surround rectangl grei size denot speed signal dynam sensor displai signal dynam sensor fig pattern result typic situat grasp object contact rest object grasp gripper tip remov object tip rest object grasp gripper center touch object side gripper user pull grasp object object start slide stabl grip slide object due grip pressur situat classifi signal distribut human observ automat classif neural network subject current research discuss equip gripper combin sensor sensor technolog lend arbitrari shape finger portion space sensor element survei control modifi number fiber bundl length materi fiber autom product moder miniatur sensor build sensit robot skin fine spatial resolut distinguish featur sensor type sensit forc low mn sensor damag deform grip forc high appli stabl grip everydai object shown gripper equip sensor suit refin grip attempt visual inform present method teach trajectori minim forc touch fiber dynam sensor research project includ automat evalu typic situat learn grasp unknown develop space combin visual tactil inform acknowledg grate jan technic bielefeld liu hong dlr static sensor share technic knowledg sensor construct robot hardwar financ neuro project german feder ministri educ research refer mark eric christoph von der mike jochen jan rolf stefan robot object percept manipul autonom eric rolf integr object recognit grasp lc intern confer intellig inform engin page ieee tactil system dlr robot hand intern symposium measur control page eric adapt flexibl robot visual tactil grasp phd physic univ ishida capacit transduc continu measur vertic foot forc med biol eng nj linear angular displac peer schmidt fur ein master physic univ decemb peer eric rolf patent zur von deutsch includ pct tabl tabl tactil sensor human skin characterist type number adapt slow slow fast fast indent proport ds dt ds dt dt constant indent locat superfici deep superfici deep recept field small larg small larg densiti variabl constant variabl constant tabl figur caption figur schema dynam sensor element capacitor ic figur circuitri dynam sensor element left part schemat middl part extract raw signal pin part combin filter circuit yield vibrat signal pin figur complet sensor static part locat central blue dynam element posit rim combin sensor mount jaw grei box attach gripper bodi process electron includ bu control figur schemat layout sensor top layer consist pressur sensit middl electrod bottom layer dynam sensor singl hair small connect surfac small hole upper layer element repeat shown figur sensor signal approach slip fast approach object approach object reveal adapt sensor signal approach speed contact place moment strong voltag diagram stop lead vibrat sensor adapt signal begin slip slip nice detect filter vibrat signal figur robot platform dof stereo camera dof tactil sensor figur translat rotat vector field hair sensor translat radial order center rotat circular order orient gripper parallel object surfac outer hair sensor serv avoid tabl object figur tactil sensor robot arm guid desir trajectori neglig forc figur exampl sensor distribut situat object contact rest object grasp gripper tip remov object tip rest object grasp gripper center touch object side gripper user pull grasp object object start slide stabl grip slide object due grip pressur figur figur figur figur figur figur figur figur figur figur 
ML/1329 6 discov fuzzi classif rule genet program roberto de alex freita julio av brazil abstract data mine consist extract knowledg data paper system discov fuzzi classif rule system evolutionari genet program algorithm evolv popul fuzzi rule set simpl evolutionari algorithm evolv popul membership function definit popul final result process fuzzi rule set set membership function definit adapt system innov idea respect encod gp individu repres rule set basic idea individu encod scheme incorpor syntact restrict facilit handl rule set disjunct normal form adapt gp oper work individu encod scheme introduct context machin learn data popular express knowledg consist rule due fact intuit human repres independ unit alter easili place content rule part part rule correspond conjunct condit verifi impli condit part rule consid true rule classic format condit constitut discret categor variabl presenc continu variabl creat situat thwart common ag problem sudden unnatur transit individu classifi safe driver todai complet year classifi safe driver lead data mine system complet predict interv singl dai altern work continu variabl overcom inconveni fuzzi logic express knowledg natur fuzzi logic flexibl power method uncertainti manag literatur techniqu discoveri fuzzi rule recent project evolutionari algorithm fuzzi rule discoveri global search state probabl converg global solut main characterist system make system system base fuzzi rule set membership function separ gener project base evolut singl popul popul fuzzi rule set evolv genet program popul membership function definit evolv simpl evolutionari algorithm system innov idea respect encod gp individu repres rule set basic idea individu encod scheme incorpor syntact restrict facilit handl rule set disjunct normal form adapt gp oper work individu encod scheme remaind paper organ section describ detail system section discuss comput result section conclud paper system fuzzi rule discoveri overview system section present overview fuzzi rule system system develop classif task data mine consist evolutionari algorithm genet program algorithm individu repres fuzzi rule set gp individu specifi pair rule condit individu rule set definit membership function interpret fuzzi rule condit individu provid popul algorithm simpl evolutionari work singl individu popul evolv principl natur select applic crossov singl individu specifi definit membership function attribut fuzzifi origin continu definit popul gp mention note categor attribut fuzzifi valu handl gp popul system simultan evolv fuzzi rule set membership function definit specif suit fuzzi rule set main advantag approach fit set membership function definit evalu fuzzi rule encod gp singl fuzzi set improv ness evalu basic idea discoveri recent main differ work system delgado work popul work delgado work genet algorithm evolv popul genet program evolv rule set work address classif task data delgado work address problem numer function approxim genet program popul rule system pittsburgh approach individu repres set rule rule condit predict predict rule attribut class valu goal attribut run system individu gp popul predict explicitli encod predict genom individu run discov rule predict singl system run number class approach process import simplifi individu avoid problem mate individu predict produc offspr individu correspond set rule encod disjunct normal form throat true ag true temperatur system function set logic oper individu repres fuzzi fuzzi version logic oper standard fuzzi oper denot membership element fuzzi set belong fuzzi set standard fuzzi set denot defin min denot minimum oper standard fuzzi set denot defin max denot maximum oper standard fuzzi set denot defin termin set consist condit attr attribut dataset attribut attr domain attr attribut attr continu mean fuzzifi system linguist linguist valu order reduc size search space order produc individu tree valid rule dnf syntact restrict tree root node except root node parent children kind node parent node children termin node parent node child node termin node condit form possibl cancel result genet termin node refer attribut rule tend produc invalid rule male sex fig show individu rule syntact constraint enforc creat individu initi popul modifi individu due action genet oper approach regard kind gp specif discoveri rule set disjunct normal make attract data mine applic fig tree repres rule main advantag work dnf directli tree convert rule set dnf gp make easier fulfil restrict hierarch posit easi collect termin node individu shown fig order check condit insert rule approach assur gp run syntact valid individu grammar implement syntact restrict drawback approach difficulti check syntact restrict lead number product rule grammar avoid logic grammar introduc complex system prefer direct implement syntact constraint select genet oper tournament select tournament size simpl individu smaller complex select complex measur formula complex number rule number condit extens motiv observ individu compet tournament fit individu individu select crossov perform similar convent gp differ case crossov oper respect syntact order crossov gener offspr crossov produc crossov oper fail children current version system crossov probabl common set literatur system offspr produc crossov necessarili insert popul popul updat strategi crossov produc offspr ad popul individu popul size individu sort fit worst individu remov popul number remov individu chosen number individu left popul constant popul set individu experi chose strategi select comparison convent strategi analog strategi emploi ea section main differ strategi top tournament classic strategi scheme system mutat oper node randomli chosen root node replac current version system individu undergo mutat probabl except individu gener undergo fit worsen fit function order calcul fit gp step comput tp number exampl cover individu rule class predict fp number exampl cover individu rule class class predict fn number exampl cover individu rule class predict tn number exampl cover individu rule class predict rule note true posit true neg correspond correct predict made individu posit neg correspond wrong predict made individu system fit gp individu comput formula tp tp fn tn fp tn data mine gener implicitli assum valu fn tn crisp commonplac assumpt invalid system discov fuzzi rule system cover rule rang correspond membership rule system comput fuzzi valu fn tn membership record rule set encod individu comput rule system comput membership condit rule membership entir rule comput fuzzi membership rule condit process repeat rule individu membership entir rule set comput fuzzi membership rule train class predict individu ls rule cover ls rule entir rule set cover current train cover rule fuzzi return membership entir rule set mean predict made individu correct wrong contribut number true posit number neg tree prune rule prune import data mine due effect code code greatli affect system perform initi prune dataset requir unaccept amount run time design oper prune gp tree basic idea oper randomli remov condit rule null coverag rule cover record cover record condit correspond remov entir rule rule set rule individu separ evalu null coverag condit drop rule condit randomli remov rule condit randomli chosen step rule remain null criterion number condit rule equal condit drop randomli rule cover record condit remov rule complet individu oper appli individu probabl individu worsen appli individu current gener motiv appli oper individu save process oper end individu undergo tree prune oper remov kind redund rule null coverag duplic rule final tree prune alter fit remov rule alter set exampl cover individu rule set membership function mention system valu continu attribut fuzzifi linguist high linguist valu defin trapezoid membership function continu attribut membership function membership function dynam modifi set paramet defin membership adapt attribut paramet membership function encod singl individu individu consid separ gp popul mention section popul gp popul individu individu divid part number attribut fuzzifi chromosom consist denot collect defin membership function shown fig gene repres attribut coordin trapezoid vertic belong pair membership function system ensur gl individu advantag reduc search space evolutionari algorithm save process number paramet optim evolutionari algorithm reduc enforc overlap membership function origin continu sum membership linguist valu intuit fig definit trapezoid membership function gene evolutionari algorithm evolv membership function perform crossov membership function evolut singl individu repres membership function result simpl evolutionari evolv mean strategi specif individu clone time clone exact copi origin individu system appli clone high rate mutat chromosom block contigu defin membership function singl probabl undergo mutat mutat question consist ad subtract small current gene effect shift coordin trapezoid vertic gene left note chromosom gene gene mutat rate chromosom correspond mutat rate gene motiv high mutat rate desir perform global search space candid membership function definit smaller mutat gene entir individu attribut modifi correspond kind local candid solut evalu fit differ solut account gene interact strategi scheme candid solut evalu differ solut effect gene modif evalu account gene interact attribut fuzzifi interact modif membership function evalu abil account attribut interact consid main motiv evolutionari local search algorithm clone undergon individu evalu fit function discuss subsect individu discard number clone experi empir determin good qualiti process time fit function recal individu popul repres definit membership function defin rule evolv gp popul qualiti individu popul depend predict accuraci individu popul scheme fit membership individu comput sum fit valu group individu gp popul comput system small portion gp popul experi report paper individu order reduc process time classifi exampl recal complet execut system gener rule set class found data set rule set classifi exampl test set test system comput membership rule set predict assign class rule set largest membership accuraci rate test set comput number correctli classifi test exampl divid total number test usual classif literatur comput result evalu system data set uci california data set repositori data set data set small number record unknown valu current version system cope record remov result report produc procedur order evalu perform system compar evolutionari system found discov crisp rule chosen comparison appli data set experi obtain good result comparison data mine system result report directli paper result produc result produc gener train test set tabl system obtain accuraci rate iri data set number bracket system standard data set system outperform system outperform iri data outperform system data set explan lower perform fuzzi rule discov system data set suggest larg number continu attribut data set suggest possibl simpl evolutionari algorithm section difficulti cope high number attribut fuzzifi case size search space larg simpl evolutionari algorithm hypothesi investig futur work tabl accuraci rate test data set system heart iri result bear mind unlik system advantag discov fuzzi tend intuit user threshold continu attribut crisp rule evolutionari system good amount comput time run singl iter process time vari coupl minut iri data set hour data set result obtain pentium ii shorter process time obtain parallel data mine techniqu point left futur research futur research system discov fuzzi classif rule system evolutionari genet program algorithm evolv popul fuzzi rule set simpl evolutionari algorithm evolv popul membership function definit tion final result process fuzzi rule set set membership function definit adapt main advantag approach fit set membership function definit evalu fuzzi rule encod gp singl fuzzi set make evalu robust order mitig problem long process system evalu set membership function definit gp individu system innov idea respect encod gp individu repres rule set basic idea individu encod scheme incorpor syntact restrict facilit handl rule set disjunct normal form adapt gp oper work individu encod scheme evalu system public domain data set compar evolutionari system found literatur data set result summar system outperform dataset equal data respect predict accuraci system competit data set data set system outperform outperform system data system advantag discov fuzzi tend intuit user crisp rule discov direct futur research gp tree prune oper system tree node prune randomli chosen research direct design prune tree node prune base estim predict power tree node note suggest research direct concern improv gp algorithm system import point investig futur research perform simpl evolutionari algorithm evolv membership function definit current version algorithm robust cope data set larg number attribut fuzzifi hypothesi investig lead improv current version simpl evolutionari algorithm refer fd genet program introduct morgan pj bentlei investig evolut fuzzi rule detect suspici insur claim proc genet evolutionari comput conf morgan la breslow dw aha simplifi survei knowledg engin mar fv modular hierarch evolutionari design fuzzi system proc genet evolutionari comput conf morgan um smyth data mine knowledg overview um fayyad al advanc knowledg discoveri data cs aa fuzzi rule induct algorithm principl data mine knowledg discoveri european conf lectur note artifici intellig aa freita survei evolutionari algorithm data mine knowledg discoveri ghosh advanc evolutionari comput aa freita sh mine larg parallel process kluwer academ dj hand construct assess classif rule john ishibuchi linguist rule extract machin learn proc genet evolutionari comput conf morgan hybrid fuzzi algorithm design compact fuzzi classif system proc ieee int conf fuzzi system ieee san usa cz genet algorithm learn machin learn gj klir yuan fuzzi set fuzzi logic wb poli ja foster evolut size shape wb oreilli pj angelin advanc genet program volum mit jj liu jt kwok extend genet rule induct algorithm proc congress evolutionari comput la usa juli dj montana strongli type genet program evolutionari comput ca sipper design breast cancer diagnost system hybrid methodolog proc ieee int conf fuzzi system se search tree build block evolutionari search classif proc congress evolutionari comput la usa juli walter ck mohan fuzzi classifi system diagnosi proc congress evolutionari comput la ca ml wong ks leung data mine grammar base genet program applic gener linguist fuzzi rule pattern classif genet algorithm principl data mine knowledg discoveri lectur note artifici intellig 
ML/2211 6 classifi unseen case miss valu zheng low school comput mathemat victoria australia depart system engin engin manag hong hong kong abstract handl miss attribut valu import issu classifi miss attribut valu train data test data affect predict accuraci learn classifi real kdd attribut miss valu common paper studi robust recent develop learn includ rel toler miss valu test data boost found similar level robust toler miss valu test data term averag error repres collect natur domain investig bag perform slightli perform perform introduct primari concern classifi learn predict accuraci recent research shown learn techniqu significantli accuraci base learn tree learn learn induc multipl individu classifi form repeat applic singl base learn algorithm classif member vote decid final classif tree learn studi classifi learn techniqu wide kdd applic handl miss attribut valu import issu classifi miss attribut valu train data test data affect predict accuraci learn classifi tree learn algorithm includ handl miss attribut valu train data test data learn techniqu tree learn algorithm base learner tree learn algorithm quinlan compar effect approach deal miss valu singl tree show combin outcom test miss test classif stage predict accuraci approach adopt approach approach deal miss attribut valu tree gener stage classif author studi carri effect miss attribut valu accuraci perform learn techniqu studi robust recent develop learn techniqu includ boost bag rel toler miss valu test data motiv research learn techniqu dramat reduc error observ obtain experi domain small amount miss attribut valu interest accuraci learn method rel test set miss attribut valu real world unseen exampl miss attribut train set complet attribut inform medic train data histor record patient attribut valu patient desir perform preliminari diagnosi result medic examin time conduct medic examin cost import classif perform unseen exampl miss attribut valu expect classif accur expect reach accuraci level attribut valu miss briefli describ idea learn techniqu section empir explor robust learn algorithm toler miss attribut valu test data section summari outlin direct futur research full version paper approach improv robust learn techniqu toler miss attribut valu test data bag gener classifi bootstrap sampl boost build classifi sequenti weight train exampl creat classifi modifi base perform previou classifi object make gener classifi concentr train exampl misclassifi previou classifi boost bag implement boost bag build classifi modifi set attribut consid distribut train set unchang attribut set select stochast combin gener tree built bootstrap sampl train data procedur combin boost calli select attribut subset adapt modifi distribut train set descript full descript found reader refer refer mention detail algorithm effect miss valu accuraci experi explor accuraci learn method rel test set miss attribut valu natur domain uci machin learn repositori stratifi carri algorithm domain miss attribut simul situat unseen exampl amount miss attribut randomli introduc miss attribut valu test set level fold learn algorithm appli train set reliabl test accuraci estim learn evalu ten corrupt version origin test set fold level miss attribut evalu result averag corrupt version deriv probabl attribut origin test set miss algorithm run train test set partit ten corrupt version test set result report algorithm domain averag trail investig studi prune tree algorithm probabilist predict vote vote decid final classif size set experi size number set result tree total probabl attribut select subset set figur show averag error rate learn algorithm domain function miss attribut level detail error rate algorithm error ratio learn algorithm discuss found observ learn algorithm significantli reduc averag error miss attribut level domain lowest averag error shown figur domain reduc comput requir experi exclud largest domain test suit partial result domain consist claim paper miss level test set error rate averag domain boost bag fig effect miss valu test data error learn algorithm averag error domain linearli growth miss attribut level boost perform bag term lower averag error rate lower averag error ratio boost test set miss attribut valu bag boost test set miss attribut valu bag perform boost toler miss attribut valu test data averag error differ boost grow slightli start drop reach averag error differ smaller averag error differ boost consid similar level robust toler miss valu test data term robust toler miss attribut valu test bag perform boost averag error differ bag grow averag differ averag differ higher perform boost perform averag error differ rel grow end point differ reduc slightli futur work empir explor robust recent develop learn algorithm rel toler miss attribut valu test data repres collect natur domain learn algorithm base learner found boost similar level robust toler miss valu test data bag perform slightli perform perform regard stochast attribut select compon contribut greater robust toler miss attribut valu test data make algorithm chanc gener tree attribut algorithm issu worth investig paper address problem toler miss attribut valu test data interest research topic robust algorithm toler miss attribut valu train data train test data acknowledg geoffrei webb kai ming ting comment research author grate ross quinlan provid research carri author visit dept system engin engin manag hong kong grate dept support visit refer machin learn research ai magazin nation confer artifici intellig menlo aaai press bag predictor machin learn boost explan effect vote method intern confer machin learn morgan kaufmann program machin learn morgan kaufmann unknown attribut valu induct intern workshop machin learn morgan kaufmann strength weak learnabl machin learn stochast attribut select australian joint confer artifici intellig springerverlag stochast attribut select multipl learn accur stabl classifi confer knowledg discoveri data mine classifi unseen case miss valu tech report school comput australia uci repositori machin learn dept inform comput scienc articl process macro packag llnc style 
IR/1530 5 search engin specif queri transform question answer abstract eugen £ columbia ny introduc method learn queri transform improv abil retriev answer question inform retriev system train stage method involv automat learn phrase featur classifi question automat gener candid queri transform train set automat evalu candid transform target inform retriev system gener search engin run question transform set perform document retriev present prototyp search appli method web search engin blind evalu set real queri web search engin log show method significantli outperform underli web search engin commerci search engin special question answer keyword web queri question inform retriev introduct number natur languag question hard submit search engin web number search servic web specif target natur languag question proprietari servic facilit interact human expert web search engin altavista googl typic treat natur languag question list term retriev document similar origin queri document answer term origin queri rank low search engin queri answer search engin recogn question question hard document queri compani websit disk storag return £ work partial author intern nec research institut summer copyright held hong kong acm steve lawrenc nec research institut nj lui gravano columbia ny search hardwar tutori glossari page definit descript hard disk good answer rigid magnet disk rotat central head store definit retriev transform origin question queri requir phrase bia search engin retriev answer document present automat learn transform natur languag question queri term phrase expect document answer question evalu set question chosen randomli excit queri compar qualiti document retriev document retriev system blind evalu relat work larg bodi research recent repres text retriev evalu confer track involv retriev short byte answer set test question work gener class answer user interest multipl answer question wai peopl part relev work retriev document collect system latest trec evalu submit origin question inform retriev system task number system aim extract answer document abnei al describ system document return smart inform retriev system process extract answer question classifi set identifi type entiti answer document tag recogn passag surround entiti correct type question rank set heurist moldovan al present system postprocess result regular inform retriev system goal return passag cardi al describ system combin statist linguist knowledg question answer emploi sophist linguist filter postprocess retriev document extract passag answer question system gener approach retriev document passag similar origin question variat standard term weight scheme passag chosen document return heurist regular express approach document similar question initi retriev user document answer document term origin question import retriev document limit number case web search engin system attempt modifi queri order improv chanc retriev answer lawrenc gile introduc specif express question transform specif phrase answer question transform phrase refer main differ current work transform hand craft set queri submit search engin differ queri syntax search engin sanderson set queri transform order retriev document descript phrase proper noun al describ system transform question hierarchi question type hierarchi built bootstrap techniqu mckeown describ experi automat build lexicon phrase collect document goal build index collect suit question answer relat larg bodi research describ method automat expand queri base relev term document interest approach present describ automat expand queri base term queri term document origin queri automat queri system expand queri run time basi initi set document return inform system origin queri contrast previou present system automat learn multipl queri optim specif search goal maxim probabl inform retriev system return document answer question exploit inher regular power natur languag transform natur languag question set effect search engin queri system suffici submit natur languag question tie search engin origin form search engin treat queri bag term retriev document similar origin queri document answer term origin queri document rank low search examin typic user page result answer natur languag approach automat reformul question queri term phrase expect document answer origin question section defin strategi transform natur languag question effect search engin queri section describ system learn reformul question section show evalu queri run time appli transform gener question phrase question train data gener candid transform answer train data evalu candid transform search engin output transform search engin figur outlin process train system question type question tabl question type phrase evalu learn transform question effect queri attempt find transform natur languag question effect queri term phrase expect document answer question learn process shown figur select question phrase stage learn process figur gener set phrase identifi categori question question categori similar goal question hard impli user definit descript hard disk goal question infer question phrase input stage set question question answer constitut train data gener potenti question phrase comput frequenc length anchor begin question result occur time output stage set question phrase quickli classifi question respect question type sampl question automat gener question train collect shown tabl method select question phrase produc includ number phrase specif wide applic stage train process limit resourc chose limit train result report phrase match regular express shown figur regular express match common concentr resourc phrase featur select natur languag process techniqu filter gener question phrase altern approach identifi categori approach number advantag approach process larg train set approach domain work languag minor modif evalu question run time categor question phrase match incorpor neglig overhead process time queri figur regular express filter automat gener question phrase candid transform gener filter candid transform stage learn algorithm figur gener candid term phrase reformul question appli filter procedur reduc comput requir stage candid transform search engin step figur candid transform gener question phrase previou learn stage procedur gener candid transform question phrase qp consist number gener initi candid transform filter phrase minimum weight filter remain phrase step detail stage learn process collect pair sampl origin collect figur describ stage learn process oper collect tag assign syntact part speech word text tagger wide natur languag process commun pair train collect match gener potenti answer phrase word prefix answer length start word boundari byte answer text sampl answer phrase gener step shown tabl phrase heavili electron comput domain phrase gener larg portion document train collect technolog relat topic phrase chang intend topic queri recal transform learn improv accuraci retriev set preserv topic origin queri filter phrase intuit good transform gener question address problem filter candid transform phrase noun observ queri noun content word express topic queri queri noun content word word noun filter candid transform phrase check gener answer phrase phrase discard part speech comput collect begin subsect result highest frequenc count appli ir techniqu term weight rank candid transform initi term weight assign candid transform appli term weight scheme term weight okapi question phrase candid transform tabl sampl candid transform gener disregard phrase noun rank formula okapi inform retriev system particip trec confer inform retriev system vector space model comput similar similar comput dot product vector repres document element vector calcul combin term weight term frequenc term document metric similar idea origin definit term document assign jone term weight respect specif queri topic calcul ö ê ö ò ö æ ò ê ö ö number relev document æ number document ê number relev ò number document weight design high term tend occur relev document smooth normal account potenti relev inform train data origin definit term weight specif queri topic appli metric task weight candid transform incorpor modif interpret queri topic question type relev document answer train collect correspond question phrase estim select candid transform respect specif question type extend term weight scheme phrase appli consist weight scheme treat singl word term comput weight candid transform comput count pair answer question match qp number relev number remain pair answer appli formula equat comput term select candid transform context select term automat queri û count term weight comput respect qp term rank strategi exploit statist relev weight aim filter assign higher weight term phrase high discriminatori measur phrase occur answer relev question type tabl phrase discrimin question question phrase candid transform û tabl sampl candid transform frequenc count term weight û result term select weight transform length candid transform tabl sampl candid transform group bucket transform length transform gener question phrase phrase occur ultim rank higher tradeoff discrimin frequenc expect explor futur work sampl output stage shown tabl candid transform sort bucket number word transform highest valu bucket expect longer phrase process differ search step order includ potenti higher transform set candid primarili shorter transform higher frequenc count chosen tabl show sampl phrase highest select weight candid transform bucket weight transform search engin final stage evalu perform candid web search engin figur show algorithm rank set candid transform singl question phrase search engin procedur repeat question phrase search engin interest step algorithm retriev set pair train exampl sort pair collect order answer pair question begin qp sort step evalu accur question shorter topic futur research set candid transform gener previou stage appli transform question time question question question remain term tion transform remov question phrase rewrit question queri candid transform question phrase pair lisp machin lisp machin comput optim run lisp appli transform question obtain rewritten queri queri lisp machin queri syntax search engin encod transform treat phrase search engin syntax queri interfac vari search engin altavista oper found oper produc significantli result preliminari experi actual queri submit altavista lisp machin googl treat term submit queri implicit semant absenc explicit oper note googl incorpor proxim queri term document rank discard word stopword list possibl rewrit requir requir part queri match combin multipl transform singl queri step figur rewritten queri queri submit search engin se top result return se retriev return document analyz step step gener step similar answer found step score count updat base similar respect detail step follow step gener document calcul accur similar measur origin answer document document retriev transform queri make assumpt answer kei phrase close proxim length overlap minim possibl answer queri document break overlap start success posit step calcul score document respect answer maximum similar extens metric modifi incorpor phrase calcul equat origin metric relev weight topic frequenc defin ã ã document length averag document length relev weight queri topic frequenc simplifi version metric trec exampl exampl candid transform queri result document result figur automat evalu effect candid transform summat includ weight learn train data previou subsect weight term phrase ø calcul û defin ø ø ø defin ø ø è ø assign procedur term encount present train collect idf document high rare low common weight deriv larger sampl million web obtain collect page trec web track case order handl phrase present train data assign weight phrase ø proport probabl term ø scale weight occurr phrase higher heurist work preliminari improv work document rank import run time oper system section result set document focu work goal rank candid transform weight highli transform tend return relev document origin document step calcul weight transform averag similar origin train answer document return transform è ë sum calcul pair set exampl result final stage train set automat rank respect effect retriev answer question match qp search engin se sampl highli rank transform qp optim altavista search engin googl search shown tabl run time queri reformul set transform automat train question store transform rule evalu question procedur figur search transform search transform engin engin altavista googl tabl top rank transform question phrase automat optim altavista googl step system determin reformul question match question prefer longer phrase prefer step set transform retriev top transform step transform rewrit origin transform result queri step transform queri submit search engin page result retriev step return document analyz score base similar document respect transform queri process score document section import differ step document retriev applic multipl final score document maximum individu document score step return document rank respect final document step top rank document return final result produc system set section present set evalu perform section detail train evalu section list retriev system comparison section introduc evalu metric perform retriev detail queri evalu relev judgment report section train collect approxim pair obtain frequent qp queri result document result maximum score document score updat maximum score document score sort document order document score return highest score question answer lisp machin comput lisp optim run lisp effici machin provid good environ program field monitor design listen field area sound speaker domin significantli indirect reflect sound bounc consol figur sampl pair train collect type pair collect tabl number train question question type question file subject figur show sampl pair obtain faq file project evalu question type number train pair collect question type shown tabl number paramet train process perform valu paramet result paramet shown tabl test paramet exhaust improv perform system retriev system compar learn queri transform specif tailor search engin evalu popular search googl altavista compar result produc system search transform evalu comparison system evalu googl googl search engin figur evalu question run time paramet descript min frequenc gener question phrase min frequenc gener candid transform max length question phrase max length answer phrase min length question phrase min length answer phrase max length prefix answer candid transform gener length document similar calcul set high includ complet answer similar calcul max number highest rank candid transform length final weight stage number pair evalu candid transform question phrase individu page timeout tabl train paramet optim googl retriev system result transform user question multipl queri transform specif learn combin queri result googl section altavista altavista search engin optim altavista retriev system result transform user question multipl queri transform specif learn combin queri result altavista section result search special answer natur languag question evalu metric inform retriev system compar base retriev document set quantifi recal document collect hand judg relev queri calcul fraction relev document document recal measur coverag system fraction relev document collect system retriev measur recal collect mark document collect relev evalu queri daunt task larg document essenti imposs billion document research address problem develop standard document collect queri relev limit domain document judg trec incorpor web track emploi collect web document rel size collect document valuabl resourc evalu inform retriev algorithm web data collect suit evalu system aim transform user queri obtain improv result exist search engin googl oper web larg evalu rigor defin altern set evalu inspect top ã page return system compar queri describ comput relev judgment document section relev evalu answer system produc metric definit ã retriev system ë queri õ percentag document relev õ top ã document return ë õ queri õ top document return googl queri judg document relev õ googl õ comput percentag question system perform system definit system ë queri õ system document cutoff ã queri õ perform system highest number relev document top ã document system note multipl system ident perform case consid queri top document return googl top document return altavista queri document relev õ top googl document top altavista googl consid perform document cutoff õ googl õ retriev system queri evalu queri relev judgment train discuss evalu perform retriev system section metric section real user question log queri receiv excit search engin portion log access consist million estim type total present evalu judg total queri total question tabl number question question type excit queri log paramet descript max number transform appli size max length queri phrase timeout individu page timeout tabl paramet evalu english question focu evalu basic question common type question found excit estim account natur languag question submit search engin set test question gener scan question phrase queri log random sampl question chosen question type sampl manual filter remov queri return document check question present verbatim train collect test question tune paramet test question type submit system modif origin question retriev url rank result url store top document document fewer retriev top document return altavista top document return googl top document return transform optim altavista paramet tabl top document return transform optim googl paramet tabl top document return special step process result page result return link question similar select editor list term question user answer page user found question similar retriev result top result bring total number retriev page assign relev judgment result url system mix present random order number volunt volunt blind system return url set evalu script present volunt result page queri page volunt judg page queri volunt told criteria judg document good page answer test question page page link link document page judg page view reason server broken result evalu result section report result evalu methodolog previou section figur show averag ã vari ã altavista googl test question optim googl highest valu document cutoff ã note perform underli search engin show larg improv altavista figur show percentag question system perform term number relev document return document cutoff test question perform question system perform poorli perform compar origin altavista search engin lower perform system perform small number comparison lower perform system metric consid possibl document return retriev simpli examin document return search engin origin queri report percentag document origin queri document examin figur figur report percentag document top æ origin queri figur plot percentag top document origin search engin figur show percentag relev document origin search engin result top document relev judgement assign figur show overlap result set retriev transform document retriev origin queri overlap low document slowli level top relev document figur averag ã altavista googl test vari number top document examin ã consist outperform underli search engin base perform system origin googl document examin experi fraction relev document retriev found underli search engin figur percentag question system return relev document vari number top document examin ã search engin altavista googl test queri highest number relev document question return multipl search engin consid return relev document lower perform system perform small number comparison lower perform system metric tabl report averag ã document set retriev separ question phrase examin document lower gener case reason document select human editor answer question note system higher top rank document question phrase figur averag percentag top relev top document top æ document return origin queri underli search engin figur averag ã altavista googl question figur report averag ã document set retriev altavista googl separ question type note system highest ã question type figur ã googl highest impress perform question due common practic web link page peopl find solv specif natur good answer type question googl exploit anchor text document achiev high result type question interest note system perform uniformli question subtyp basic question type explain phenomenon observ question question fall question typic express request definit explan term request list characterist featur concept volunt judg complain question hardest evalu answer test question creat direct creat free html web host web design softwar manual clear judg page relev type question note perform type question suggest approach rout question search perform question type transform question manner optim search engin note small number question categori limit accuraci result individu question type futur work summari avenu futur research improv system exist method extract passag document implement domain natur languag techniqu improv identif question type multipl transform combin singl queri question rout search engin perform question type multipl search engin simultan plan investig creat phrase transform content word question plan make transform process dynam transform expect high submit base system lower transform fall back origin queri introduc method learn queri transform improv abil retriev answer question inform retriev system method involv classifi question question gener candid queri transform train set question question system type phrase ã aj av tabl averag ã altavista googl separ question type question ã document evalu candid transform target inform retriev system techniqu process natur languag question applic wide rang inform retriev system implement evalu method appli web search engin blind evalu set real queri web search engin show method significantli outperform underli web search engin common question type acknowledg author acknowledg nec research institut bulk research support nation scienc foundat grant robin burk give train collect faq file format nick craswel give document frequenc count collect million web page eric glover regina kathi anonym comment final great peopl time busi schedul evalu regina nicola david john eric tobia sharon eugen carl roman michael brian anonym refer singhal answer extract real world implement answer extract intern workshop expert natur languag inform system brill simpl part speech tagger confer appli natur languag process brin page anatomi hypertextu web search engin intern world wide web confer inform retriev text aaai fall symposium ai applic knowledg navig page bucklei examin role statist linguist knowledg sourc system experi textual question answer harman result challeng web search evalu intern world wide web confer sanderson retriev descript phrase larg amount free text cikm lawrenc gile context page analysi improv web search ieee internet bucklei improv automat queri acm ru tool surf answer net robertson term select queri journal volum page robertson jone relev weight search term journal american societi inform robertson walker relev weight relev inform sigir beaulieu okapi automat ad vlc interact track salton automat text retriev inform comput mckeown experi autom lexicon build text search elicit queri excit web search engin cikm overview eighth text retriev confer question answer track report xu croft improv effect inform retriev local context analysi acm transact inform system 
ML/447 6 machin colour constanc good brian lindsai martin school comput simon fraser british columbia canada abstract paper present neg current machin colour constanc algorithm good object recognit result algorithm successfulli correct colour balanc imag displai colour balanc typic applic colour rare put comput vision goal show method obviou machin colour vision object recognit colour constanc method test prove insuffici import find addit present result show correl colour constanc perform object recognit colour constanc recognit rate introduct set show machin colour constanc matur point aspect machin vision colour constanc algorithm requir fraction practic long result suffici accur beg repres task provid answer question simultan give measur accuraci past perform colour constanc algorithm report term averag angular error rm error predict target imag test colour constanc put object recognit object recognit strategi swain method base compar histogram distribut imag colour colour index fail ambient light illumin object recogn differ construct model imag swain ballard suggest colour constanc preprocess address solv introduc rel colour absolut colour colour histogram colour constanc method work natur task preprocess imag prior index swain ballard origin suggest fact colour index sensit variat ambient scene illumin expect method scene illumin directli affect imag rgb colour question address exist colour constanc algorithm effect gener colour descriptor colour index work typic rang scene illumin encount practic fluoresc offic light goal test colour develop improv colour index modif outlin paper colour index import colour constanc method colour correct good estim illumin descript colour constanc method neural setup descript test result discuss colour index colour constanc task machin colour constanc algorithm gener descriptor scene colour measur term camera rgb coordin camera output affect surfac reflect illumin red channel spectral sensit red channel equat green blue channel spectrum incid spectral reflect surfac assum rel spectral power distribut spatial invari intens colour constanc method develop exploit spatial variat illumin surfac colour chosen illumin colour descriptor machin colour constanc problem express deriv imag scene canon illumin rgb canon imag scene rgb unknown unknown illumin map account chang rel spectral power distribut unknown canon illumin rgb space defin output soni colour video camera strictli speak human observ paper refer rgb colour constanc method estim chromat colour canon illumin ignor intens compon wai normal rgb elimin effect intens colour index base standard chromat coordin colour correct algorithm discuss colour constanc algorithm convert chromat word chromat unknown illumin rg unknown provid estim chromat rg canon canon illumin colour index perform chromat histogram swain ballard major test rgb includ test space method simpl model histogram creat imag object system recogn object separ background built segment manual imag object chromat histogram determin unlik case model test object separ imag background test histogram intersect model histogram intersect defin model highest histogram intersect score identifi unknown object implement chromat histogram sampl larg imag sampl toler inaccuraci colour constanc colour constanc colour correct test colour constanc neural network algorithm estim colour incid illumin estim transform imag colour canon colour case estim transform directli test algorithm previou test shown perform colour constanc colour correct step case base diagon model illumin chang diagon model von kri adapt coeffici rule diagon model simpli state effect move scene illumin model scale channel independ scale factor scale factor written element diagon matrix previou work shown diagon model work full linear model typic scene illumin type sensor found video narrow band sensit diagon model work algorithm work chromat space case diagon transform model chang chromat move illumin diagon model chromat coordin defin equat show choic chromat coordin crucial diagon model illumin chang present origin coordin colour constanc algorithm test briefli turn version algorithm compar averag rgb imag ideal grei rgb grei rgb grei canon diagon scale factor colour correct simpli grei averag grei averag grei averag algorithm compar rgb white canon maximum found imag band separ variant version correspond case differ al maximum colour channel diagon scale factor colour correct simpli canon max canon max canon max studi shown method perform machin colour constanc method method deriv constraint ambient illumin evalu differ gamut colour found imag canon gamut experi construct canon gamut hundr reflect spectra wide varieti common object canon gamut convex hull set rgb valu arisen reflect illumin canon illumin understand rgb tripl imag scene unknown illumin presenc reveal canon gamut repres full set expect spot canon illumin correspond rgb insid canon gamut obtain illumin canon longer lie canon gamut set diagon transform map back canon gamut repres set unknown illumin intersect possibl figur solid line show convex hull diagon map transform chromat vertic canon gamut dash line show convex hull map canon gamut map describ unknown illumin restrict intersect convex set simpl reflect result chromat canon illumin canon gamut defin convex hull happen hull intens case elimin move chromat coordin defin equat imag rgb convert chromat coordin turn lie canon gamut map canon correspond repres canon map requir scale compon compon correspond canon gamut point case scale canon gamut point yield scale pair figur plot map fill diamond case correspond point insid convex hull canon gamut linear scale map interior gamut point result scale interior convex hull map figur convex hull set repres complet set map canon gamut point convex hull figur repres hypothesi unknown illumin point model chang creat move canon illumin unknown illumin convex hull figur express constraint find imag unknown illumin illumin repres point convex hull illumin possibl result colour canon gamut rgb yield constraint unknown illumin yield constraint chromat map hull vertic canon gamut shown open circl figur map scene point lit unknown unknown illumin repres map intersect convex hull candid illumin elimin convex hull set distinct imag call imag gamut vertex imag gamut yield constraint unknown illumin intersect constraint obtain vertic map constraint set establish heurist method pick remain candid estim unknown illumin hull centroid final estim method possibl illumin constraint result report base method ad illumin constraint measur lot light sourc reveal restrict gamut sampl illumin includ measur illumin includ indoor outdoor illumin light includ concret build light filter illumin unusu exclud result illumin gamut reformul term set diagon transform map illumin canon illumin form intersect constraint imag gamut constrain estim unknown illumin method carri chromat space rgb space method case constraint set polyhedr convex hull potenti case estim surfac bright addit surfac experi requir bright run method convert back chromat space note carri method convert result carri place method estim final estim made maxim volum intersect set method final estim origin chromat estim hull centroid case neural network colour constanc report result shown good perform neural network colour constanc network estim illumin chromat base gamut colour present imag neural network perceptron hidden layer input layer consist binari input repres chromat present scene hidden layer smaller neuron output layer neuron imag rgb scene transform standard space discret present input layer input binari indic presenc absenc chromat imag output layer neural network produc valu chromat illumin output valu real number rang train network algorithm thousand synthet imag gener randomli select illumin reflect surfac reflect integr spectral sensit function camera accord equat learn phase network provid imag data chromat illumin test imag imag experi colour object show pictur soni colour video camera balanc light gamma correct turn essenti linear function lumin rgb camera calibr apertur set pixel clip band apertur manner left imag dark overcom problem extend dynam rang imag roughli averag frame store result float point record imag produc versatil imag simul effect imag apertur effect clip bright spot imag illumin top section part light judg ii light booth illumin judg ii illumin cool white philip judg ii judg ii fluoresc full blue produc illumin similar colour temperatur deep blue sky effect creat chang illumin figur ball illumin illumin spectra plot figur set imag imag object move object group regist imag set object move imag set imag evalu colour index scene illumin object posit imag figur object imag singl illumin figur illumin spectra illumin filter cool white figur spectra test illumin measur adjust camera calibr model subtract variat background averag background determin averag larg number imag len cap adjust pixel darker amount tabl make small camera subtract intercept camera linear data imag result imag closer model rgb valu simpli integr equat incom spectra multipli sensor sensit function comput imag data clean averag block pixel exclud dark pixel recognit perform colour constanc error rm error rgb map canon clip clip rough fit figur object recognit perform function colour constanc error result experi confirm obviou hypothesi colour constanc improv colour index situat illumin test object construct model figur show clear correl colour index perform colour constanc error recognit perform measur figur base weight averag colour rank recognit colour index calcul match strength model strongest match fact correct rank match correct object rank obtain percentag total match rank distil result singl repres weight sum percentag weight rank weight rank weight rank match rank consid failur count measur colour constanc error base comparison correct imag set regist model comput root squar differ basi entir imag chromat target canon colour correct imag adequ canon illumin illumin creat blue cast imag red green intens low possibl reliabl result colour constanc algorithm obtain colour constanc averag produc data plot fill diamond figur note data set extend dynam rang imag order prevent clip visual check colleagu suggest exposur level normal view averag time higher simul imag captur level scale threshold greater match result plot fill triangl figur part motiv experi clip level unexpectedli good perform method compar neural net method compar result report al expect special natur imag clip failur reli pixel clip pixel accur specular highlight provid excel clue colour provid clip imag gener specular preserv maxima channel perform usual human user automat apertur control adjust captur process obtain invari clip pixel case start break maximum channel bad data point verifi degrad simul clip level result plot figur clear clip level algorithm degrad quickli clip level scale perform close report result perform algorithm case shown tabl rm map error clip level artifici clip level extend dynam rang imag white patch grei world gamut map neural net actual illumin diagon map figur rm map error versu artifici clip level rm map error rm differ chromat correct imag target imag curv explicitli depend clip rm differ fewer data point method degrad faster algorithm algorithm rm error actual neural net tabl colour constanc error error rm differ chromat map imag regist target imag result averag estim map imag illumin denot colour constanc simpli input imag output imag denot true unknown illumin rgb colour correct base diagon model rank perfect actual grei world gamut gamut neural net tabl number match rank percent total number attempt match match make case test illumin canon illumin exact origin question machin colour constanc good base feel answer tabl show index perform colour index algorithm preprocess perform find object result rank base actual illumin rgb colour correct colour index diagon model illumin chang support perform conclud distanc machin colour constanc task support object recognit term feel gave machin colour constanc reason result consid closer illumin spatial uniform chromat care remov tempor spatial averag troubl develop good camera model requir algorithm avoid bad match due colour background place object black cloth aspect experi open data real imag object random everydai object object includ object fluoresc compon box fluoresc compon chang matt compon present problem diagon model imag research assist understand intend elimin unintend bia choic object test illumin repres dramat rang term encount common natur light standard light situat ad unusu effect thing heavili filter theater light test machin colour constanc algorithm comput vision task object recognit base colour histogram intersect colour constanc algorithm perform expect base previou result colour balanc imag expect colour constanc process provid colour descriptor accur colour index perform close obtain chang ambient illumin shown tabl turn case colour constanc yield improv simpli improv figur show improv histogram match linearli relat rm error colour predict result al human colour constanc accur state result repres complet constanc complet absenc result experi question human colour constanc suffici accur object current state machin colour constanc perform challeng futur research clear rm error colour predict typic vision task object recognit base absolut descriptor rel colour refer bv learn color fourth color imag nov int comp color constant color ieee tran patt anal mach color recognit object illumin invari properti color opt soc nov colour constanc scene spectral vari fourth european confer comput vol page april method recov surfac spectral opt soc colour constanc vari proc intl conf comp jun algorithm color intl journal comput vision color gener diagon transform opt soc color pami oct vol number ba analysi theori color journal optic societi america oct quantit studi vision vol pp colour theori msc simon fraser school comput rg introduct theori neural publish learn intern error propag de rumelhart jl mcclelland parallel distribut explor microstructur cognit foundat mit color constanc natur imag asymmetr journal optic societi america sept 
DB/2761 3 web log data wareh mine intellig web cach pedreschi institut italian nat research council dipartimento di di pisa abstract introduc intellig web cach algorithm emploi predict model web gener idea extend lru polici web proxi server make web access model extract web log data data mine techniqu approach studi frequent pattern tree result algorithm show substanti improv exist cach term hit rate design develop prototyp support data wareh web log extract data mine model simul web cach algorithm kei web log data data frequent associ tree introduct data mine aim discov regular pattern hidden emerg area web mine aim discov regular pattern structur content web web resourc access paper describ mine applic base data develop intellig web cach capabl adapt behavior basi access pattern usag extract histor access data record log mean data mine techniqu author email preprint submit elsevi scienc april idea extend lru recent cach replac polici adopt web proxi server make web access extract web log data introduc wai construct intellig web cach algorithm emploi predict model web request goal algorithm maxim hit percentag request web entiti retriev directli request back origin server gener idea motiv observ lru polici drop cach recent entiti base assumpt request occur recent past occur futur assumpt true explain lru request tempor local case web request inform extract histori recent inform cach replac strategi clear indic mine web log data access model emploi compar altern variat lru cach present approach uniqu adapt usag natur web due fact cach strategi parametr wrt data mine recomput period order track recent past adopt data mine yield class web cach frequent pattern tree rst extract past web log pattern form web pattern mean request user session pattern cach replac algorithm assign high prioriti due lru treat analog develop basi web log model capabl request web object time futur request properti predict base histor data web log prerequisit techniqu develop data log raw log le unsatisfactori start point data mine develop fulli autom acquisit process migrat log data carefulli design data orient analysi requir intellig web cach data web log inform select prepar data mine deriv attribut miss valu attribut approxim nal step knowledg discoveri design refer web cach model mean evalu model extract data mine emul parametr replac support evalu comparison replac select hit rate weight hit account size request entiti log data acquisit model extract evalu web cach formal implement manag microsoft sql server prototyp system rst round extens experi larg log le web server present result scope benchmark pursu perform gure obtain develop compar substanti hit outperform metric tradit cach strategi adapt data de nite motiv studi subject intellig web cach web cach web cach inherit techniqu issu cach processor memori le system peculiar web give rise number issu call adequ solut cach deploi di erent classi ed depend cach locat network spectrum possibl rang cach close client proxi server cach close origin server server proxi server cach interfac client server ectiv proxi cach reli common access user set object closer client browser perform cach basi local le system singl user activ web server cach interfac singl web server user reduc number request server load scalabl avail recogn deploi cach world wide web improv net trac wai includ reduct bandwidth network latenc server load web pose issu risk reduc applic ectiv mention issu cach consist copi dynam object web resourc person secur legal issu order evalu qualiti web cach measur depend resourc bottleneck system commonli criteria basic hit ratio request ful lled handl origin server straight measur save workload cpu web import web server comput resourc main issu rough estim save network trac reduct latenc perceiv client weight hit ratio byte serv client cach measur network trac save web relev network bandwidth web server cach issu time wait retriev resourc short latenc simpli strongli resourc multimedia strict requir work evalu cach strategi simul set collect log web server abstract architectur detail estim latenc restrict hit rate weight hit rate cach algorithm replac consist order object cach paramet arriv request object composit object evict order replac strategi mention fifo arriv size object number lru request number request request cach make room object size rst evict object size greater lru object size section strategi refer call inform time extend strategi weight entiti number request receiv access object entiti higher valu size remov rst observ schema simpl estim strategi simul histor futur request order comput algorithm found similar extend schema strategi refer theoret complex analysi cach strategi notic approach nement static follow xed criteria origin priori analysi gener behavior request trac yield data independ account structur characterist actual request ow move web context di erent characterist web proxi stai context behavior perform static strategi signi cantli chang explain approach paper adapt make mine model modi ed time time recent request histori cach refer model intellig cach section introduc gener model cach strategi data mine algorithm integr replac polici introduc give present cach speci cation hypertext transfer protocol http protocol protocol client send request server request method access resourc identi er messag request modi er client inform http method includ client request request process request store client request resourc server respond request return entiti content protocol consid time provid origin server estim cach consist evalu consist entiti cach expir entiti consid exact copi entiti store origin server expir time entiti cach stale order cach stale entiti client valid valid perform mean condit messag origin server intermedi includ time resourc origin server check entiti chang tradit cach strategi model entiti cach belong prioriti queue cach replac strategi coincid prioriti queue weight assign polici tradit cach strategi weight assign polici xed claim strategi adapt ow possibl ect weight assign polici call polici exploit knowledg extract past request weight assign polici xed boil tradit cach strategi gener model intellig cach strategi report figur initi data mine model built past data long hit loop forev hit client retriev fig intellig cach refer model request entiti cach entiti fresh entiti return client perform measur case consid hit byte back client count weight hit rate measur note size entiti necessarili number byte back client di client close connect receiv entiti weight entiti possibl weight element updat weight assign polici express entiti cach miss entiti delet cach fresh version retriev push cach push method consist assign weight entiti updat weight element cach make cach maximum entiti pop cach weight approach run replac polici time cach size ow data mine model period updat condit xed time interv cach perform model updat method notic updat model decoupl cach perform extern detail real world cach abstract gener model request entiti entiti abstract assum request refer document perform hit rate weight hit refer percentag entiti request client found cach assum log le includ inform assum transact data mart refer document abstract cach consist assum transact data mart includ size entiti time request mean entiti cach consid fresh size coincid size entiti time request assum entiti size valid section data mart web log data develop data mart web log speci calli support intellig cach strategi data mart popul start web log data raw server log le assum basic eld data mart popul consist number preprocess code step perform data clean transform data mart implement relat microsoft sql server process popul data mart formal autom framework log le produc server text le row http transact assum row common log le format includ eld accomplish http standard ip address client comput origin method uniform resourc locat resourc request http protocol http return code mean size byte client worth note assum set basic server addit inform transact log http header describ transmiss paramet content properti resourc techniqu adopt provid deep inform structur resourc incom outcom relev content classi cation interest categori fig data mart tabl relat worth investig work restrict evalu approach assumpt deep inform obtain data mart mine log data data mart implement relat schema shown figur main tabl data mart weblog row tabl store transact server describ eld uniqu identi er transact insert id uniqu identi er transact respect tempor order transact webserv identi es real server transact prog uniqu identi er order transact transact time inform maintain order transact relat hash code identifi request compon host paramet separ hash code discuss size number byte client size request entiti time request basic eld raw le log comput section load balanc server implement network comput real switch rout client request fig load packag legend symbol sql server dt packag numer code le extens request url depth server le system request resourc numer code top directori pathnam request resourc code imag remain eld suit data mine analysi deep tabl decod hash valu text decod numer valu hash function assign code url mean url consid queri data mart phase data mine distinct pair con ict url store order evalu power hash code worth hash function demonstr sucient decod tabl tabl result unnecessari data mart popul process data prepar data mart popul design sql server data transform servic tool process data text applic task de ned dt packag shown figur central task packag perl script perform preprocess raw le log assumpt section select transact request method client request http return statu mean request url dynam gener entiti consid http transact method consid omit transact transact involv partial success error select url rst split eld inform code integ hash function standard syntax http url ab path server ab path absolut path request resourc server ab path form path queri collect pass input resourc execut cgi script distinguish site server path rel le system server request le extens paramet preprocess log set normal perform url order distinguish syntact distinct url refer resourc url code trace back problem ineci manag system data mine algorithm work string advantag numer mention optim disk ecient comparison privaci concern addit code mechan emploi implement intellig cach cient code url independ code url reason eld paramet normal code hash function bit integ hash function obtain xor bit code return hash algorithm entiti gener dynam resourc url cooki approxim section assum avail size request entiti time request eld basic log data construct eld heurist transact data assign maximum size previou request url transact request record discard data mart assum entiti size client receiv heurist correctli assign eld client close connect receiv assign correct entiti size previou request client receiv entiti assum entiti size time heurist assign incorrect entiti size request request form approxim unavoid cach consist algorithm estim expir time entiti posit assign uniqu identi er id transact order registr log le prog real web server identi er note assum xed order arriv request real web server interleav addit data mart eld addit eld weblog tabl includ inform suit data mine algorithm eld count long past futur url request count consid number request note set larg constant previou request data mart set case access data mart eld count number access url paramet de nabl data mart user xed denot frequenc request url previou request class eld freeli de nabl data mart sql updat queri data mine algorithm store classi cation transact distanc msec request previou request user set sucient larg constant case rst request user uniqu integ identi er user session transact belong concept user session deserv detail discuss semant point user session de ned set number request entiti rank berkelei cumul frequenc request byte fig distribut request entiti rank distribut request byte transmit entiti size url access user imposs distinguish access object inform content restrict heurist xed break user user session heurist maxim forward session identi ed maximum number link user backward refer made adopt refer length base assumpt amount time user spend examin object relat interest user object content model user session built distinguish navig object link interest content object inform user distinct navig content access relat distanc request access time delai greater threshold consid content navig url user session sequenc navig url content url start assign url request group request distanc threshold note identifi user ip address heurist fail case ip correspond dynam comput share user correct identi cation user presenc form authent login workload experi extens literatur understand statist characterist workload server impact cach polici introduc workload show statist distribut ect typic characterist report refer cumul frequenc request berkelei number request page delai berkelei fig distribut request work williamson popul data mart data web dai log data lead italian portal web total million request total transmit ip distinct entiti normal total web server size berkelei juli log data depart electr engin comput scienc california berkelei total million request total transmit ip distinct total web server size hash code algorithm gave rise ict invari observ williamson con rm predomin request html multimedia document small median transfer size kbyte kbyte small fraction distinct resourc normal request total number request signi percentag resourc request cite paper observ concentr request refer small fraction url ect distribut shown figur refer entiti request entiti distribut follow zipf number request entiti paramet williamson observ cumul frequenc request similar observ hold cumul frequenc byte distribut characterist present shown figur workload entiti size byte byte account hit rate cach size lru size fifo weight hit rate cach size lru size fifo fig hit weight hit rate wrt cach size byte transmit local refer request resourc small amount time con rmed workload figur show distribut request case time request url occur request user session mention presenc proxi distinct user session singl session show ip predomin number request berkelei workload top ip count total number request figur relat berkelei distribut request time distanc request previou client plot ts distribut number request paramet con rm expect user access navig deploi refer model data mart data present section instanti gener intellig cach model figur rst base frequent tree introduct gener mine task descript applic cach summari result overview data mine model process result simul obtain build simul extend lru workload introduc section workload split dataset train set build data mine valid set valid extract model simul extend lru strategi valid set cach size mbyte mbyte server server exponenti growth figur show hit weigh hit rate tradit cach strategi valid set berkelei notic strategi tend limit hit rate cach size grow limit maximum hit rate weight hit rate achiev sucient larg cach berkelei workload limit hit rate limit weight hit rate frequent pattern discoveri introduc notion frequent pattern de ned discuss intellig web cach form frequent associ correl item user transact discov user transact user sequenti tempor correl item user transact discov user transact set item user associ rule rule set support de ned percentag user transact item con denc de ned ratio percentag transact item probabilist approxim probabl item rule approxim probabl item transact item number associ rule built set transact grow exponenti number rule support con denc greater threshold call minimum support minimum con consid applic item correspond web user transact correspond user session rule inform mean user expect possibl order necessarili consecut associ rule account time compon adapt sequenti pattern discoveri context user session sequenc request resourc user sequenti pattern express form support sequenti de ned percentag user note experi cycl model figur session subsequ associ interest frequent sequenti pattern support greater minimum threshold experi associ rule sequenti pattern restrict due concern design cach strategi implement ecient sequenti pattern length form associ form frequent intellig cach lru strategi essenti consist assign request entiti prioriti counter increment approach lru criteria assign strategi extend modifi prioriti entiti cach reaction incom request set sequenti pattern extract train past log data request resourc request set extract predict resourc request suggest evict accomplish assign prioriti request immedi strategi minimum support local sequenti number pattern global larg number pattern high support substanti modi cation lru impact pattern quanti ed number pattern quanti cation number resourc uenc problem manag larg number pattern hash tabl ecient access pattern examin bucket hash tabl repres point list pattern url note number pattern larg lead larg number weight resourc cach order bound adopt prune extract pattern interest comparison approach associ rule sequenti pattern extract associ rule train modifi prioriti resourc cach request rule set extract rule rule subsum con denc sequenti sequenc subsequ exist km hit rate cach size pattern ass rule lru weight hit rate cach size pattern ass rule lru fig hit rate weight hit rate wrt cach size pattern compar evalu tempor correl improv pure correl result workload split dai train set dai valid rst dai log data extract sequenti pattern user minimum support threshold rang session maximum equal result extend lru strategi simul remain dai log data figur show result obtain simul lru extend lru extend lru figur show hit rate curv similar converg cach size grow graph reveal signi hit improv approach standard lru cach rang absolut gain cach size largest figur show analog plot weight hit rate improv smaller absolut due fact approach maxim account size cach entiti hit rate extens weight hit rate similar mean tempor aspect plai relev role extract pattern berkelei workload show di erent behavior split dai train set dai valid set network trac dai smaller previou workload larger number larger time window reconstruct user session signi rest absolut gain absolut improv stand di erenc hit rate hit rate cach size pattern ass rule lru weight hit rate cach size pattern ass rule lru fig hit rate weight hit rate wrt cach size hit rate cach size lru hit rate cach size lru fig berkelei hit rate wrt minimum support length di erent threshold minut yield reason session factor uenc result berkelei workload larg percentag request origin proxi potenti larg set real client hidden sequenc request di erent distinguish client mix session longer request cover request mislead pattern perform order avoid maximum session length session longer ned threshold request broken shorter lower support threshold workload order extract number rule train set figur show result simul obtain maximum absolut gain hit weight hit rate strategi yield hit rate signi cantli higher cach medium size di erent enhanc strategi perform poorli berkelei workload cach intuit behavior resourc request pattern cach action probabl cach workload distinct case berkelei intellig strategi rare weight cach size probabl cach intellig strategi weight worth investig global approach figur show perform scale respect pattern minimum support plot figur support tune order obtain sucient larg number pattern lower support valu yield perform minimum support extrem strategi turn worst perform tree recal notion predict classi cation model base show enhanc lru strategi classi er model describ discret call entiti term attribut call observ attribut classi er construct set object train class valu predict unknown class valu object set popular classi cation tree algorithm tree tree data structur consist node leav node speci es test observ attribut outcom child node present path root leaf tree base observ attribut valu object class speci ed leaf class predict tree context notion class intellig observ eld distanc access section implement cach strategi recal assign weight entiti product distanc request entiti calcul histor central idea section train classi form predict chosen class variabl continu discret valu class classi intellig cach order tree data mine model instanti gener refer requir make choic discret method appli set observ attribut construct tree integr cach strategi weight assign polici present select call strategi gener rule de ning class order scale cach size intuit intend approxim train tree weight assign formula tightli relat behavior train set cach size cach size strategi run request train averag entiti size number time maximum number entiti note valu comput train set eld strategi simul discret choic mean decid number class discret valu interv experi result small number class turn prefer strategi discret de ned class de ned class de ned class de ned class de ned intuit choic intend approxim predict successfulli cach resourc request probabl depend number entiti cach hold cach size interv time cach hold train entiti assum request interv extrem case intermedi request refer distinct mean possibl entiti remain cach refer possibl reduc interv reduc interv simul strategi assign higher prioriti request lower class valu weight assign intuit approxim adopt weight assign function correct lru weight ad displac relat prioriti request strategi lru weight function request de ned lru assign weight request entiti hit rate cach size lru weight hit rate cach size lru fig hit weight hit rate wrt cach size low weight mean low prioriti prioriti entiti time proport number request receiv request entiti proport entiti de ne weight function proport function alia approxim mean class predict tree approxim assign predict class correct discret adopt assign give class assign class formula relat fact request class higher probabl successfulli cach request class note express byte lru assign function measur request number request free byte de ne worth observ weight assign entiti chang refer make implement depart weight request observ attribut set observ attribut tree built time url request attribut hit rate cach size lru weight hit rate cach size lru fig hit weight hit rate wrt cach size hit rate cach size poor rich weight hit rate cach size poor rich fig hit weight hit rate wrt cach size select strategi basic eld hour part date continu basic eld discret result figur report perform strategi lru term hit rate weight hit rate valid set workload split set approxim equal mention di erent split valid set half yield similar result hit rate gure similar exhibit impress improv wrt consist absolut improv simul achiev gain lru consist gain lru theoret strategi note achiev hit rate outperform tradit strategi report figur weight hit rate perform di erent workload perform improv tradit strategi figur consist gain wrt modest gain wrt berkelei tribut behavior lower predict accuraci tree berkelei workload correctli classi ed request valid compar workload correctli classi ed natur limit approach strategi partli answer question investig relev tree assumpt paramet xed discret weight formula variant strategi poor strategi tree assign class perform distinct strategi tree built observ attribut strategi tree built observ attribut number request url previou request experi valu similar result cach size rich strategi tree assign correct class entiti belong strategi show predict accuraci classi er adopt necessarili mean strategi exhibit perform term hit rate weight hit rate due role paramet de nition strategi figur show perform strategi berkelei workload hit note higher predict accuraci mean higher perform cach size lower worth note improv accuraci tree classi cation result perform impli act paramet strategi order improv hit rate weight hit rate cach higher predict accuraci mean higher perform strategi substanti mean addit inform eld frequenc contribut improv perform nal comment pentium iii pc hard berkelei workload cach size strategi hour tree produc store memori depth prototyp search tree time present approach enhanc web server cach data mine model pattern built ical data design suitabl data mart main problem design solv perform gure develop compar lru side theoret strategi substanti hit rate strategi outperform tradit xed strategi hit rate weight hit rate extens benchmark requir proxi server combin approach extens tradit cach extens dynam strong indic data cach systemat enhanc hit weight hit rate intellig web cach strategi li capabl tune cach replac algorithm recent histori request extract web log xed albeit extrem ecient bound diculti pattern web usag acknowledg research partial support fst grant mine project acknowledg support microsoft grant grate team fst give opportun conduct discuss refer cr ea cach limit proc inter web confer cc jl ps cach polici web technic report research cc aggarw ps disk cach web object proxi proc agraw fast algorithm mine associ rule larg proc intl confer larg acm comput survei mf perform evalu web proxi cach replac proc intl confer comput perform model techniqu tool volum lectur note comput scienc mf cl internet web workload perform transact network uni ed approach approxim resourc alloc proc annual acm symp theori comput web client access characterist cach world wide web la studi replac algorithm virtual storag ibm system journal analysi navig behaviour web site integr multipl inform vldb journal web cach evid proc confer cao maintain strong cach consist world wide ieee transact comput cao www proxi cach proc usenix symposium internet technolog system activ cach dynam content proc ifip intl confer distribut system platform open distribut process ms js ps data mine pattern web proc intl confer distribut comput system group web page refer transact mine world wide web proc ieee knowledg data engin exchang workshop data prepar mine world wide web knowledg inform system discoveri interest usag pattern web proc web usag analysi user pro ling workshop volum lectur note comput scienc bd web cach content deliveri bd survei proxi cach evalu proc intl web cach workshop depart electr engin comput scienc berkelei cs http mf improv proxi cach analysi replac ieee internet comput measur access characterist client proxi proc usenix symposium internet technolog system gold commun acm han data concept techniqu san internet data bradlei sarawagi acm sigkdd explor jin web cach exploit sourc tempor local web request proc intl web cach content deliveri workshop jin algorithm web proc intl confer distribut comput system jin tempor local web request cach implic proc sigmetr confer kp wareh mine web proc acm cikm workshop web inform data manag web mine acm sigkdd explor ep main memori cach web comput network isdn system microsoft microsoft sql server han discov web access pattern trend appli olap data mine technolog web proc advanc digit librari conf je pitkow mm simpl robust cach algorithm base dynam access proc intern www confer program machin learn san technic report rfc internet engin task forc rizzo replac polici proxi transact network ecient ieee tran knowledg data engin spiliopoul lc web util workshop web data base volum lectur note comput scienc web usag discoveri applic web usag pattern web acm sigkdd explor uniform resourc identi er gener syntax technic report rfc internet engin task forc hypertext transfer protocol technic report rfc internet societi log control cr ea remov polici network cach web proc acm sigcomm august rr collabor web cach base proxi proc sigmetr confer 
HCI/975 4 anthropomorph agent spatial languag multimedia technolog system bielefeld abstract paper describ commun virtual environ main emphasi process spatial express natur languag instruct work part project interior design domain multiag system act intellig mediat user graphic system make commun spatial relat develop anthropomorph agent graphic visual scene gure explain qualit spatial introduct interact graphic system user concentr imagin free technic import improv interact virtual environ intuit commun form work interfac intellig mediat translat qualit express natur languag quantit command graphic system aim process verbal spatial express develop articul anthropomorph agent visibl instanc agent encourag natur languag conceiv cation user agent place user ey virtual environ commun situat spatial instruct paper concentr process spatial express interest simul human visibl robot control natur problem vari perspect import smooth exibl di erent perspect natur languag deictic frame view addit frame consid recent present approach comput empir evalu mean basic spatial relat space main emphasi clarifi depend distanc shape respect simpl ideal object fig anthropomorph agent virtual ce room project complex set visibl spatial issu realist gure kind gestur possibl simul natur index spatial express compar ed addit problem dealt select actual frame refer section rst explain system section describ commun space presenc anthropomorph posit adverb discuss idea give outlook futur work section project environ project research focu theme cial intellig comput bielefeld goal develop intellig form commun virtual environ applic chose interact design explor manipul scene detail research project partli support ministri scienc research feder state grant augment graphic db object hierarchi geometr scene descript bookkeep graphic system hamilton control virtual camera refer color space coordin control mechan input data structur parser type listen speech listen gestur listen observ verbal gestur input fig architectur system commun system natur languag set agent altogeth form multimedia user translat qualit instruct human user quantit technic command updat visual scene model instruct issu multimod input agenc combin input di erent modal consist input coordin listen type speech gestur track analyz sensor data data coordin analyz integr input receiv listen gener intern task descript post agent hamilton agenc mediat agent system track exact object locat negoti altern wai act space agent comput spatial transform virtual environ scale scene object inspect modifi color agent identifi object mean color descript red chang object camera agent calcul transform virtual camera enabl navig scene resolv ambigu refer qualit refer agent determin rank list candid refer scene object hamilton control agent realiz manipul articul gure bookkeep agent author access modifi augment graphic data base suppli current situat inform agent request agent realiz mean instanc slightli vari function visual scene anthropomorph move chang wai rotat agent move horizont plane turn vertic axi graviti resist solid bodi account user deal agent experi acquir physic world head agent rotat rotat vertic axi angl rotat horizont axi maximum optic restrict avoid user orient virtual environ short period time head turn back special frame refer head import current system point gestur object agent improv commun gestur implement gestur arm extend index nger stretch comparison fig form bodi gestur easi grasp commun agent turn view object idl time arm turn back automat chang size possibl bodi size explor exibl provid instruct shrink grow agent adapt individu size user includ applic demand addit point implement wave gestur answer input agent turn virtual wave arm action evok answer wave user simpl data glove chang perspect system switch extern view view place virtual camera forehead hamilton virtual camera posit part hamilton cap visibl cover screen user possibl explor virtual extern view agent visibl scene serv anchor frame bodi user direct agent move scene assess ergonom featur furnish size comparison featur fig hamilton involv view user adopt perspect eld vision anthropomorph scene fig show view hamilton point gestur commun space everydai life awar commun space di cult describ rout direct furnish room aspect structur di erenc space space languag base time aspect speaker listen di erent point view clari ed actual frame refer posit orient time solv problem contextu knowledg gestur deictic refer front import part dialogu focu topic subsect discuss applic interact graphic system type locat deixi posit deixi deixi deixi term direct depend posit orient speaker listen percept dimens determin biolog physic graviti asymmetri human bodi human bodi symmetr direct refer system descript spatial frame refer implicitli utter understood listen determin frame di erent coordin system chosen speaker deictic origin posit listen person ax determin human percept space addit bodi special frame refer head upper part import posit orient human bodi speaker make refer refer object orient refer object everydai front side desk side peopl sit depend front left ax structur wai call face align modal utter speaker explicitli mention frame refer posit express speaker awar ambigu stand face face chang left situat knowledg actual understand ambigu utter empir nding select frame refer depend geometr factor distanc user refer optic factor contextu factor situat applic impress frame fig set experi refer chosen actual carri simpl experi random subject subject stand distanc meter angl front desk locat subject instruct move fig user deictic hamilton deictic view left move posit turn subject pot posit pot posit tabl show fact signi variat prefer subject inclin judg orient tabl uenc subject posit subject appar chose posit deictic perspect case conclud solut observ individu di erenc account fig bowl desk user deictic hamilton deictic view applic system di erent frame refer hamilton direct move point view extern viewpoint user depend frame translat space carri di virtual agent user face face fig show possibl realiz instruct left transform object user fulli awar frame refer fig show realiz instruct move bowl point view featur bowl move posit deictic view move posit deictic view user realiz object move posit deeper understand frame evalu relev literatur similar experi result vagu contradictori miller conclud frame refer easier deictic consid descript furnish room peopl deictic point view postul static situat frame refer frequent studi show signi prefer deictic point set select actual frame refer highli uenc contextu factor explor peopl spatial perspect actual imaginari listen human perspect research speaker dont share viewpoint neutral perspect studi result show import interact feedback speaker primari goal understood reason speaker perspect perspect interest featur frequent neutral descript requir person perspect minim partner ort partner highli perspect open question optim relationship system user addit gener perfect express creat perfect build architectur accept understand repair observ choic frame refer mean perspect relev make varieti contextu factor consid individu percept situat individu languag import extract gener rule exibl account frame refer realiz instanc space agent calcul transform object instanc object view deictic view user deictic view hamilton hamilton agent calcul movement anthropomorph gure realiz deictic frame refer mention evalu system rst carri transform deictic point view realiz meet expect correct system state system gener solut di erent agent instanc comput transform base di erent instanc work deal adapt individu user prefer advantag realiz improv robust system instanc agent carri task unoccupi space goal instanc activ possibl solut futur simpl hand gestur data resolv ambigu refer fig instruct combin hand gestur movement hamilton point view posit deixi english languag adverb posit deixi posit space depend posit speaker vari interpret adverb di cult relev frame refer complic di erent place time speak listen origin coordin system move gestur verbal refer abstract place region expand di erent found region includ place speak region exclud place speak german languag mean refer shorter wider distanc speaker region interpret posit adverb system commun space restrict simpl instruct transform object move furnitur virtual user refer region visibl room region visibl room abstract region relev limit context make describ interpret adverb suggest system correct user interact chair deictic refer user posit di erent perspect user assum clari ed user extern user posit formal virtual camera determin current eld vision user view anthropomorph agent identi es posit agent user involv frame refer instruct mention chair move anthropomorph agent virtual front screen region displac gestur region user direct point arm desk verbal express combin di erent form facial point arm nger system anthropomorph agent carri point gestur arm subsequ express region direct point arm complement point gestur issu user data glove describ region object space easier point gestur issu agent visibl refer posit anthropomorph agent mean listen easili imagin interpret bit move desk clue locat posit line view user involv user move virtual room hamilton bit move desk larg part interpret instruct vagu real speaker give short hint head ey graphic kind gestur di cult understand support current system realiz comput instruct inform di erent agent bookkeep knowledg hamilton agent spatial knowledg comput goal posit instanc hamilton agent refer di erent frame refer instruct agent instanc bookkeep inform bookkeep decid adverb refer refer object gestur posit camera hamilton relev mention gestur gestur perspect gestur perspect gestur gestur perspect gestur perspect relev hamilton realiz task object current adverb repres posit extend region actual goal posit object move determin object close comput constrain detect futur work desir gestur interact discuss futur work spatial dialogu increasingli attract attent di erent research area paper present anthropomorph agent graphic system add comfort respect spatial languag introduc visibl illustr possibl virtual room carri point wave gestur psycholog reason gure encourag natur main interest improv situat spatial commun deictic investig adverb di erent frame refer dealt instanc agent realiz instanc expect user actual refer frame account visual system posit adverb percept gure instruct user refer posit orient anthropomorph agent point gestur agent posit object result present realiz process spatial express requir larg amount situat knowledg peopl pronoun vagu descript distanc posit visibl angl perceptu context context instruct current instruct gestur refer object frame refer direct posit fig contextu factor experi psycholog factor gener context di erent refer object qualit posit direct work possibl regard uenc contextu factor give fuller account contextu factor area futur research center fig show spatial aspect frequent ambigu arrow uenc divid gener context import process spatial express inform automat adapt system individu user prefer refer die der fischer und zur und deixi im und ch empir valid model comput spatial al advanc arti cial zur bei bericht nr au dem sfb und im contrast studi english klein studi deixi demonstr benjamin wo ist bericht learn adapt interfac agenc distribut arti cial intellig meet machin learn environ springer zur de und auf pn languag percept cambridg automat depict spatial nation confer cial intellig view spatial preposit ai magazin mf ect spatial perspect choic work note process spatial ijcai montreal interact graphic design situat agent graphic springer linguist strategi nativ articl process latex macro packag llnc style 
ML/594 6 knowledg model framework learn adapt knowledg wolfgang ralph bergmann kaiserslautern centr learn system applic depart comput scienc po box germani abstract paper present framework learn adapt knowledg knowledg light approach reason system knowledg light mean approach acquir knowledg insid cbr system describ sourc knowledg insid cbr system knowledg present framework term knowledg appli framework case studi knowledg light approach learn adapt knowledg point issu address design algorithm learn adapt knowledg point view issu topic research final close short discuss outlook work introduct investig learn adapt knowledg approach learn adapt knowledg found dial chef system knowledg intens deriv analog approach veloso learn adapt knowledg knowledg intens mean approach requir lot background problem solv knowledg dial chef adapt strategi special problem field acquir base gener domain knowledg reduct knowledg acquisit cost necessarili case knowledg engin effort costli work focu call knowledg light approach learn adapt knowledg knowledg light mean term learn adapt knowledg knowledg light approach dont presum lot knowledg acquisit work acquir knowledg insid system learn adapt knowledg learn paramet adapt acquir knowledg approach learn retriev featur weight learn algorithm vsm nn vsm learn featur adapt weight hill climb algorithm possibl learn global adapt weight local weight prototyp case weight adapt determin influenc featur target complex work area induct learn algorithm extract adapt knowledg case case base paper categor knowledg insid knowledg base sketch framework learn adapt knowledg start point guidelin design adapt learn algorithm earli start point methodolog knowledg model adapt learn approach focu knowledg light knowledg elicit effort knowledg model task learn paramet adapt acquir knowledg approach learn retriev featur weight learn algorithm vsm nn vsm sourc knowledg cbr system richter althoff richter wilk cbr system store knowledg knowledg mean domain knowledg problem solv knowledg describ domain knowledg insid mention vocabulari describ case similar measur retriev solut transform adapt hold knowledg includ compil knowledg develop time actual problem gener includ human knowledg engin activ case base consist case specif knowledg interpret run process problem solv compil manual compil human acquisit mainten task difficult system gener interpret knowledg acquisit mainten task potenti easier requir updat case base part attract cbr flexibl pragmat decid includ knowledg compil versu case interpret develop cbr system gener aim manual compil knowledg absolut cbr system develop decid knowledg distribut depend avail knowledg engin effort case knowledg fill knowledg transform transform found improv similar measur learn knowledg transfer case base similar knowledg transfer adapt knowledg light approach learn adapt knowledg knowledg code focu learn adapt knowledg adapt improv adapt acquir adapt knowledg provid gener framework make compar evalu approach problem develop ad case case base compil happen maintain cbr system learn design made problem solut discuss workshop framework learn adapt knowledg chapter describ framework learn adapt knowledg light discuss knowledg learn adapt knowledg knowledg figur abstract view process learn adapt knowledg induct algorithm sourc knowledg previou knowledg similar case base adapt knowledg transform adapt knowledg learn algorithm vocabulari measur improv adapt knowledg adapt case base induct learn algorithm preprocessor exampl fig adapt knowledg learn knowledg focu induct algorithm gener gener knowledg exampl cbr system consist exampl model real world problem incomplet select knowledg learn depend kind select knowledg induct algorithm data preprocess suitabl result set exampl set attribut attribut deriv knowledg final integr output attribut descript learn exampl induct algorithm learn process adapt knowledg form adapt improv knowledg case studi appli framework rule learn approach mention introduct approach learn adapt knowledg kean algorithm build pair case featur differ case pair build adapt rule briefli describ algorithm term preprocessor build pair case extend note featur differ target differ case inform base kean suggest constrain case pair limit number advantag similar measur select pair suitabl learn input learn algorithm case pair comput preprocessor step learn algorithm build adapt rule case pair featur differ precondit target differ rule subsequ refin gener extend coverag rule base complex gener perform adapt time kean describ learn rule constrain guid explicitli domain knowledg manual compil adapt automat rule learn place kind inform preprocessor adapt sourc input dynam part adapt consist set adapt rule addit inform confid rate rule confid rate reliabl rule inform calcul learn algorithm base gener appli gener rule algorithm control rule applic strategi resolv conflict dynam part adapt describ approach integr learn rule rule learn determin gener rule improv confid rate discuss direct main point paper provid start point framework learn adapt knowledg knowledg light approach complex adapt knowledg hard acquir techniqu number problem approach classif task numer symbol target learn paramet sophist adapt solut learn algorithm learn paramet adapt acquir knowledg approach learn retriev featur weight learn algorithm vsm nn vsm complex work area induct learn algorithm extract adapt knowledg case base algorithm build pair case featur differ case pair build adapt rule acknowledg author prof michael richter discuss suggest work fund commiss european commun contract ii inform knowledg reengin reason irish medic system kaiserslautern bibliographi intern confer algorithm learn theori call paper richter wilk althoff richter wilk technolog experienc base construct forthcom deriv theori reconstruct problem solv acquisit machin artifici intellig volum page morgan lo ca leak learn adapt strategi introspect reason memori search david reason volum page menlo ca aaai press symbol classif algorithm case studi version space european workshop lectur note artifici page springer verlag learn anticip avoid plan problem explan failur annual nation confer artifici intellig page usa morgan kaufmann publish view plan memori task academ massachusett learn adapt rule case diploma thesi triniti dublin learn adapt rule case base advanc reason page springer verlag case adapt incomplet causal model reason research page springer verlag expert case base learn adapt prior case annual florida artifici intellig research page combin rule case learn case adapt seventeenth annual confer cognit scienc societi similar metric learn classifi neural knowledg similar measur invit talk scale machin case studi deriv analog prodigi machin learn method chapter page morgan san mateo weight featur reason research page springer 
IR/2245 5 maximum cliqu problem larg graph gc resend abstract present approach cliqu comput larg discuss graph decomposit scheme break problem piec manag dimens greedi random adapt search procedur find approxim solut maximum cliqu problem maximum problem larg graph present experi heurist real data set collect telecommun industri graph order million vertic edg introduct massiv data set bring seri special comput challeng data set model larg special set edg attribut repres special characterist applic hand understand structur underli essenti storag organ inform retriev paper present approach find larg cliqu larg million vertic edg undirect graph vn set vertic set edg subset denot subgraph induc graph complet vertic cliqu subset induc graph complet maximum cliqu problem find cliqu maximum cardin graph undirect graph integ multipl edg defin similarli maximum cliqu problem complex approxim remain open question recent papadimitri yannakaki introduc complex class max snp show natur problem complet rel reduc preserv qualiti approxim vertex cover problem constant minimum cut domin set max problem complet problem solut complet problem approxim arbitrari small constant optim solut problem class approxim arbitrarili small constant factor question octob kei word phrase maximum cliqu larg data larg applic lab research technic tr resend approxim scheme found complet problem class left unresolv berman show max snp problem polynomi time approxim maximum cliqu problem approxim polynomi time perform ratio maximum cliqu size approxim number vertic graph al connect approxim complex interact proof system breakthrough approxim complex recent result al shown maximum number satisfi formula approxim arbitrari small constant resolv open question immedi show find good approxim solut list problem shown polynomi time algorithm approxim maximum cliqu size factor np result al complex result worst case maximum cliqu problem problem solv maximum cliqu problem approach exact solut heurist approxim problem expect exact solut method limit perform larg problem heurist close actual maximum cliqu solut main heurist approxim maximum cliqu problem local optimum global optimum handl heurist devic escap poor local optim solut devic present simul anneal tabu search genet algorithm move solut heurist grasp sampl region solut find local minimum time inform algorithm heurist maximum cliqu problem practic applic scienc engin project classif fault code comput inform signal transmiss theori align dna protein sequenc paper organ section graph decomposit scheme experi section grasp find larg cliqu larg graph preliminari comput result section conclud remark made section graph decomposit discuss decomposit scheme make larg graph suitabl process graph optim algorithm discuss scheme approach partit edg set underli direct graph edg max cliqu larg graph underli undirect graph edg vertex defin subset vertic neighborhood respect nc respect deg nc preprocess extern memori algorithm comput connect compon connect induc vertex set classifi vertic sourc sink transmitt partit direct edg set connect compon comput direct depth search tree sourc vertex high order edg greedi random adapt search procedur tailor local maximum cliqu problem section collect defin neighborhood structur search improv local cliqu direct weight vertex edg run collect weight direct edg worth notic suggest topolog order process undirect cliqu give global structur inform product maximum cliqu size time maximum local cliqu size upper bound maximum cliqu size experi heurist neighborhood structur move compar local max cliqu glu search improv simpl decomposit scheme work larg connect compon graph approach repeatedli appli reduct size graph reduct assum cliqu size smaller interest edg vertex delet simplif reappli reduct scheme section illustr reduct scheme graph real telecommun data grasp maximum cliqu maximum smith greedi random adapt search procedur maximum independ set problem tailor maximum cliqu problem maximum problem larg base procedur grasp iter method construct greedili solut find local optim solut neighborhood construct solut role random gener solut initi local search greedi lead resend procedur set initi cliqu set subgraph induc vertic deg respect deg select random nc end end figur grasp construct procedur construct solut good qualiti local search appli greedili solut converg quickli local minimum probe potenti neighborhood carri limit amount time initi describ grasp maximum cliqu describ construct mechan local search procedur construct phase grasp maximum cliqu build vertex time vertex guid construct assum cliqu construct step construct phase cliqu hand call vertic cliqu select vertic vertex candid includ cliqu adjac select cliqu vertic denot set candid vertic vertic restrict candid list candid vertic high subgraph induc candid vertic vertex high graph induc candid vertic deg respect subgraph induc candid node real paramet interv select random ad cliqu construct select newli chosen vertex candid vertic adjac elimin set candid candid set simpli neighborhood newli select vertex respect candid set process repeat set candid vertic empti figur show construct phase grasp local search implement wai simpl approach seek vertex cliqu remov adjac vertic cliqu includ cliqu size pair vertic cliqu remov max cliqu larg graph procedur adjac vertic select adjac vertic end end figur grasp local search procedur procedur select interv end end end figur grasp maximum cliqu problem permit vertic cliqu cliqu figur show local search implement figur show indic procedur make grasp maximum cliqu grasp section requir access edg vertic graph limit graph small fit memori procedur work vertex subset edg edg secondari disk storag enabl smaller memori procedur describ speed comput grasp describ defin oper graph paramet delet vertic incid edg procedur cliqu sampl subset edg tg tg function graph subgraph denot vertex set procedur grasp appli produc cliqu size cliqu found lower bound largest cliqu vertex resend procedur subgraph tg subgraph tg end partit tg set vertic end max tg end end figur approach maximum cliqu possibl maximum cliqu discard appli paramet procedur grasp reappli reduct aim delet irrelev vertic grasp focu subgraph interest reduc size graph grasp explor portion solut space greater grasp iter faster smaller graph reduct result subgraph smaller grasp made explor solut space detail number iter graph repeat applic grasp reduc graph desir size partit edg remain set smaller appli grasp result subgraph size largest cliqu found paramet oper reduct achiev procedur cliqu call figur show approach procedur edg graph sampl discuss seek find cliqu connect examin compon time maintain edg share vertic close sampl cliqu edg select comput depth search tree max cliqu larg graph direct subgraph compon store edg sampl order determin tree relat problem interest find larg graph defin subgraph cliqu complet defin subgraph vertic edg densiti number edg defin optim problem exampl max fix max fix max approach find cliqu extend address problem grasp construct cliqu serv seed grow modifi local search procedur construct procedur ident case cliqu local search procedur vertic adjac vertic vertic remov ad size note local search procedur ident cliqu procedur similar find larg cliqu deriv procedur sampl edg origin graph subgraph induc vertic edg size grasp appli subgraph produc size origin vertex greater discard possibl size experi larg graph outlin preliminari exploratori experi sampl dataset experi silicon graphic challeng comput mip processor gbyte main substanti amount disk space current data telecommun vertic million edg found connect compon compon size greater pair giant compon vertic detect suggest wit behavior similar predict random graph theori graph random giant compon depth search tree giant vertic vertic interest tree size induc subgraph time log subgraph vertic argu largest cliqu compon size greater cliqu subgraph induc vertic resend tabl cliqu found construct local cliqu found distinct size construct local cliqu distribut expect occur larg largest million edg count edg observ remain edg tree cliqu graph induc node tree largest subgraph induc vertic tree expect cliqu larger found exploratori phase integr grasp procedur procedur appli manual begin consid edg larg compon remov vertic appli result graph vertic fit memori graph search larg cliqu motiv identifi lower bound size maximum cliqu delet vertic larger portion graph possibl identifi larger cliqu grasp repeat iter produc local optim cliqu appli local search construct solut run time point appli local search construct solut explor improv cliqu size independ natur grasp iter comput configur creat independ run grasp start random number gener seed tabl summar part result cliqu size number grasp iter construct improv size number distinct cliqu found grasp iter interest observ share larg number vertic appli greedi procedur cliqu identifi disjoint set cliqu produc cliqu size cliqu size size max cliqu larg graph freq cliqu size figur frequenc cliqu size found data consid edg larg compon remov vertic result graph vertic edg iter grasp produc cliqu size cliqu size produc cliqu size produc cliqu size produc figur show frequenc cliqu size found algorithm found cliqu size quarter intensifi search entir huge connect compon remov vertic result graph vertic edg figur show frequenc cliqu size found algorithm figur show statist improv attain local search grasp iter carri vertex edg graph cliqu vertic found cliqu optim expect cliqu larger vertic found local search improv construct solut larg construct small cliqu construct cliqu size improv local search size final confid cliqu size found appli result graph vertic thousand edg ran grasp iter graph parallel processor half dai finish largest cliqu found vertic cliqu share vertic resend freq cliqu size figur frequenc cliqu size found entir dataset comput test larg densiti paramet size found conclud remark present approach cliqu comput larg discuss graph decomposit scheme break problem piec manag dimens greedi random adapt search procedur find approxim solut maximum cliqu problem larg graph present experi heurist real data set collect telecommun industri graph order million vertic edg varieti industri applic deal explicitli edg multipl extend method present find direct weight subgraph satisfi special local densiti condit make amen local search base techniqu find direct report forthcom paper refer dynam model massiv data set technic lab florham decemb function extern memori algorithm european symposium volum lectur note comput page intract approxim problem technic max cliqu larg graph local min construct solut figur local search improv approxim maximum cliqu technic berman complex approxim independ set problem lectur note comput resend approxim maximum cliqu proc ieee symp foundat comput page ta resend probabilist heurist set cover problem oper research ta resend greedi random adapt search procedur journal global ta sh smith greedi random adapt search procedur maximum independ set oper glover tabu search part orsa glover tabu search part ii orsa test case gener comput result maximum cliqu problem journal global holland adapt natur artifici system michigan ann johnson editor graph dimac implement volum dimac seri discret mathemat theoret comput scienc american mathemat vecchi optim simul anneal papadimitri yannakaki approxim complex class proc twentieth annual acm page pardalo maximum cliqu problem journal global page ta smith fortran subroutin approxim solut maximum independ set problem grasp technic lab florham acm tran math softwar yannakaki approxim maximum satisfi proc annual symp discret network servic research lab shannon florham nj usa center appli depart industri system fl usa inform scienc lab florham nj usa 
IR/1486 5 open hypermedia lesli carr sean bechhof carol gobl wendi hall multimedia depart electron comput southampton uk keyword open link metadata word count abstract inform manag group depart comput scienc manchest oxford road manchest uk paper describ attempt project defin deploi conceptu open hypermedia servic consist ontolog reason servic repres sophist conceptu model document term open hypermedia link servic offer rang facil scalabl integr form conceptu hypermedia system enabl document link metadata describ content improv consist breadth link www document retriev time reader author time author creat concept metadata metadata data describ data enhanc librari catalogu schema canon exampl metadata fall broad catalogu artist structur pictur colour semant person hold joi metadata activ major focu interest www inform publish digit librari extens markup languag concern applic express data recent defin resourc descript framework aim rdf provid standard framework express statement data statement give inform version keyword inform attribut dublin core metadata link provid conceptu inform attribut web page import enabl search engin provid queri result pertin concept simpl keyword hypermedia system distribut link servic make inform provid rudimentari cluster document tag keyword retriev link document tag navig keyword effect classifi document cluster share set variat stem achiev kind associ requir web document link dimens base content construct link manual inconsist obfusc chief reason associ content similar conceptu hypermedia system hypertext structur behaviour term conceptu schema type document includ conceptu domain model describ document content inform hypertext repres explicitli metadata reason domain model classif structur classifi document share metadata deem similar author link document activ author concept link document link open hypermedia system link servic common usag web involv link document html web consid hypermedia system inher web infrastructur prevent hypertext link abstract document manag link open hypermedia system link class store manag separ multimedia document cach instrument research hypermedia commun increasingli web publish applic adopt open hypermedia approach dl power framework aid navig author address issu distribut inform manag intermediari model dl add link annot document deliv proxi origin www server ultim client browser number softwar modul opportun ad kind link creat navig overlai coher interfac set unlink resourc archiv address open citat project dl treat link creation resolut servic provid number link resolut engin resolv peopl bibliograph citat potenti link anchor heurist knowledg base link resolv hardwir monolith system chain sequenti document link ad previou resolv inher synchron arrang mean delai delai critic path document process tightli coupl distribut link resolut servic link resolv distribut multipl server decoupl deliveri document aim complex involv implement conceptu inferenc logic provid ad document author deliveri core document queri system ad portfolio link resolv provid consist navig link base concept content multimedia page unlink document parser ontolog link link servic servic gener request editori vocabulari knowledg request resourc instanc servic link document request hint metadata servic figur ontolog servic map natur languag term concept graph resourc servic obtain web page repres concept link gener ontolog term make link editori knowledg prune expand link ontolog semant link servic integr browser consist java applet monitor interact browser set javascript function manipul html dom compon link servic brought bear web page receiv longer form obstacl deliveri document set link page refresh ontolog servic java shown oper figur manag set concept relat schema ontolog form concept relat store xml queri result mediat simpl xml document type figur ontolog servic queri term concept relationship link gener modul link servic contact ontolog servic obtain complet list languag term repres concept ontolog term occur gener ontolog servic prefer prefer term map concept identifi concept string link gener contact resourc servic obtain list document instanc concept number destin identifi link anchor editori modul evalu number qualiti potenti link obtain gener number link consist format request broader narrow term ontolog servic order expand cull set anchor destin term document constructor add hypertext link present style behaviour figur show link ad document control panel figur link behaviour shown debug link expand document metadata servic independ servlet document decor languag term specif ontolog servic harvest specif tag document appli extern document independ effect declar rang process specif region content correspond term ontolog figur page figur link cloth figur term link correct number destin figur prune part link resolut process editori knowledg compon advantag implicit structur ontolog make inform kind link make select set list link usefulli reduc whilst broaden concept strategi number link altern approach concept predefin ontolog candid anchor creat link section list system concern manipul wai repres mode deploi meta tag tag author inform web resourc highli uncontrol tag unconstrain term varieti page special thing applic metadata tightli bound document order discov metadata examin document central metadata process web robot harvest cach metadata web resourc resourc figur simpl meta tag index duplic metadata easi access web issu mainten crucial difficult index date repositori index simpl gener ontolog list arbitrari keyword yellow page yellow page page classifi content taxonomi hierarchi subject area broken gener achiev classif hierarchi categor page manual page situat metadata store extern document real link classif web resourc resourc figur yellow page yellow page problem classif date autom approach gear support human locat resourc provid knowledg shoe simpl html ontolog extens develop parallel understand system group depart comput scienc maryland shoe mechan definit ontolog assert claim resourc respect ontolog assert web page includ page mark html base meta tag inform agent page shoe assert form instanc assert membership class relationship instanc web ontolog resourc resourc ontolog figur shoe queri engin shoe shown figur inform spread metadata attach explicitli document shoe html gather place robot queri browser user examin metadata situ reach shoe annot page ontobrok ontobrok system architectur similar wai shoe annot web page kb ontolog metadata express framework specif annot data queri ontobrok shoe share characterist annot document reli web crawler harvest store knowledg base kb queri ontolog schema queri form differ coupl aspect shoe ontolog extens mechan explicitli place ontolog web clear gain access ontobrok ontolog make link instanc markup ontolog appli ontolog describ content document multimedia repositori metadata construct author present fit user metadata index emphasi ontolog order structur result close relat ontolog produc structur rdf rdf descript differ system shoe ontobrok framework implement system system ontobrok rdf format compar rdf framework talk metadata rdf data model base idea relationship intent rdf metadata held separ document attribut separ metadata document rdf rdf document repositori access web resourc resourc rdf document figur rdf ontolog search engin figur show rdf descript work note rdf simpli number wai thing put rdf prescript ontolog specif rdf order minimum data set dublin core clear ontolog sit web rdf document form repositori collect metadata number resourc discuss system focu primarili metadata inform discoveri intend make metadata annot order build construct hypertext discuss issu relev issu discoveri versu author conclud describ space metadata applic base rich model system close system link associ resourc fix origin author tag html open system addit extra structur resourc document necessarili direct manipul amend origin proxi similar add link metadata resourc read approach metadata provid metadata resourc discoveri resourc annot metadata describ content user queri metadata order find present result open hypermedia concern resourc author process construct hypertext link ad metadata close world manner repositori metadata exist queri order discov find resourc discoveri close metadata thing resourc locat ad metadata framework resourc describ link resourc open metadata resourc figur show benefit ad metadata approach close metadata link resourc open metadata induc link resourc close discoveri open figur ad metadata discoveri author author resourc discoveri implicitli target link obtain metadata prototyp differ system support extra inform link exist web page note confin set web page link servic deal web long mechan metadata document involv match word term recognit explicit metadata resourc ontolog simpl richer model emploi stage project hypertext approach conceptu metadata web concern resourc discoveri user locat resourc inform web treat huge hypertext structur link resourc locat base content markup indic line link search blur gener resourc discoveri approach sit firmli search end spectrum contrast intend address problem construct hypertext build simpli provid resourc discoveri metadata annot discov author system agent locat resourc annot result simpli present rank list metadata index mechan link metadata author return notion link ad sourc document result metadata annot resourc determin metadata ascrib determin relev simpl index process retriev resourc index question approach involv insert metadata document repositori kind queri space metadata approach section focu issu metadata annot web resourc question metadata resid examin exist approach introduc space compar system term distribut addit materi notion express metadata figur attempt classifi metadata approach fit relat open open web internet cd filesystem dl platt altavista wai ontobrok keyword thesauri ontolog dl tambi dwq metadata author provis structur author descript space read richer queri figur metadata space dimens improv repres rich metadata move keyword ontolog final underli reason descript repres open distribut intent move simpl standalon file system distribut addit link final open hypermedia classif thesauru static hierarchi order index web resourc platt demonstr applic sit web open hypermedia point link gener runtim metadata model simpl keyword tambi system suppli access distribut inform sourc net descript logic model dwq wareh project descript logic approach problem improv schema integr distribut link servic keyword anchor ad link document dimens improv gain addit metadata structur shown figur metadata rich author provid descript turn enabl reader pose richer queri meet inform searcher addit structur author power jump open hypermedia architectur enabl link author provis metadata holi grail upper hand corner graph open hypermedia richli express model reason futur work ontolog structur space languag term consist keyword descript difficult creat subsequ lead model concept inaccur document interpret commun develop control base thesauru languag art architectur thesauru wordnet gener languag thesauru base semant net acm comput classif system base abstract hierarch classif scheme classif scheme hamper larg static unsound singl classif hierarchi resembl base systemat ontolog declar specif concept principl linguist term phrase rigor fix interpret attribut human interpret lack make hard queri check extend principl reason base terminolog resembl collect elementari concept assembl form complex composit concept sound effect terminolog repres knowledg scheme express support dynam automat classif complex composit concept base scheme descript logic convent frame semant network logic concept subsumpt satisfi reason servic offer dl flexibl construct evolv conceptu network retriev stage develop ontolog servic incorpor dl control updat mainten ontolog encount term web system conceptu model form link document explicitli navig classif scheme implicit issu present render visibl ontolog link queri construct inform navig impli inform user potenti link document share relat concept user search term control vocabulari control constraint visibl ontolog link queri construct issu investig final editori activ link servic strengthen concept document calcul link concept spread evenli document space ontolog space strategi link cull valid instanc effect kind action depend design construct conclud remark aim conceptu hypermedia servic escap limit pure lexic string match approach link discoveri prototyp success link construct softwar interact independ suit servic fact offer reason altern link candid select link unsuit reason demonstr advantag approach subsequ work improv user interfac link improv sophist ontolog process mainten descript logic acknowledg work support epsrc grant refer transpar access multipl bioinformat inform sourc intellig system molecular biologi canada pp place produc manipul web seventh www brisban support public art galleri collect intern confer workshop expert system applic dexa septemb uk classif base navig pictur archiv ifip confer data kluwer wd terminolog terminolog server inform environ proc ieee conf softwar technolog experi pd model retriev document index express sigir forum pd aid search acm link queri intern project de open inform comput network isdn elsevi de distribut link tool author world wide web journal oreilli associ comput classif system current statu futur mainten report cc updat comput jan conceptu navig world wide acm hypertext june acm san texa ontobrok research advanc technolog digit librari springer verlag lnc xml link languag version candid recommend juli integr link resolut servic queri open hypermedia workshop hall hill distribut hypermedia link ieee comput societi press creation hypertext link document measur retriev effect journal american societi inform fellbaum electron lexic mit isbn access inform sourc world confer www internet hawaii open hypermedia control metadata web intern world wide web confer browser proxi support open hypermedia structur mechan www eighth intern world wide web cope chang ontolog distribut aaai confer ontolog manag workshop aaai press knowledg languag internet applic dept comput maryland colleg technic report lagoz harnad develop servic open integr impact link acm confer digit librari concept base design data dwq acm intern confer manag data dalla usa resourc descript framework model syntax specif recommend hypertext engin approach wilei son structur type incorpor knowledg acm hypertext flag taxonomi open hypermedia acm hypertext introduct art architectur dublin core warwick review literatur march septemb automat construct manag larg open inform process navig automat link base semant close inform process manag 
Agents/997 1 investig interact agent agent control compon thoma brett victor ping xuan depart comput scienc massachusett amherst umass comput scienc technic report februari abstract explor agent context agent coordin research intellectu question major issu pertain interact agent chosen class explicit model interact address interact paper ambiti attempt address frame context coordin research introduct base long histori work agent agent control compon build distribut ai attempt frame address set intellectu question pertain agent interact li heart issu interact agent possibl occur level interact machineri hold agent underli machineri control individu agent henceforth term coordin protocol describ specif dialogu agent held coordin instanti protocol coordin denot larger group concern agent reason plan resolv carri commun activ return issu coordin mechan section notion mechan intertwin intellectu version paper workshop polici effort sponsor advanc research project agenc air forc research laboratori air forc materiel agreement number nation scienc foundat grant number number govern author reproduc distribut reprint notwithstand copyright annot thereon view author interpret necessarili repres offici polici express advanc research project agenc air forc research nation scienc govern ffl assum model agent engag multipl ramif interact interact account level underli agent control agent engag dialogu agent attempt task mutual task contract contract recoveri gener place decommit possibl penalti gener trigger agent control machineri ffl held coordin multipl agent gener entail exchang task goal inform tempor constraint inform view bind variabl machineri machineri queri oracl truth mainten agent bind provid result time multipl candid task negoti requir underli agent control mechan identifi constrain situat part interact involv queri underli oracl oracl inform order make prioriti accomplish result commit intend impact question interfac machineri lower level control ffl situat approach issu perspect ff agent hard deadlin loom lack suffici time coordin soft task interact coordin modul machineri reflect upcom deadlin option includ end exist dialogu refrain start modifi reflect hast case involv simpli termin standard requir dialogu includ branch tempor requir anytim problem neat cyclic inform obtain commun influenc agent belief impact intent plan action continu dialogu gain inform chang choic agent made result agent time time spent simpli detract domain problem solv question address issu implic machineri argu agent address issu address attempt frame question lead implic agent pertain matter deal activ level abstract discuss section interact conjunct interact agent control compon push harder issu interact question outcom effort modifi agent coordin gpgp support situat adapt issu clear resourc simpl object requir network perform action resourc degrad base state resourc degrad vari plicat domain current project interfac agent control technolog process view task send robot team hazard environ perform unman explor damag build access structur applic requir protocol behavior applic coordin agent intellig environ inform gather agent effort open gpgp applic adapt redesign reimplement import concept gpgp creat gpgp import note view agent control differ perspect agent question pose relev agent technolog question role agent research work coordin hand work structur dialogu agent formal implic inform exchang coordin work gener pertain make agent area research relat benefit cross fertil explor research context work akin begun abstract underli agent machineri simplifi agent task model addit context requir properli frame understand question interact agent machineri section describ view agent control domain independ architectur discuss machin approach coordin protocol specif return question pose section agent control compon approach agent control problem domain independ perspect domain problem process program sophist problem coupl domain independ task model taem modul agent coordin agent schedul possibl compon learn diagnosi problem solver translat intern taem structur pass control compon larger prototyp agent architectur shown figur control compon focu paper shown figur taem task model taem environ domain independ task model framework describ reason complex problem solv process taem model coordin research research collabor distribut design distribut situat assess surviv system intellig environ hospit patient schedul coordin softwar process typic problem solver repres domain problem solv action possibl level pass taem model agent control problem solver coordin modul schedul taem model hierarch abstract problem solv process describ altern wai accomplish desir repres major task major interact process translat transform abstract process program taem task structur schedul coordin task assessor execut subsystem execut monitor problem solver current goal organ organiz design problem solver detect diagnosi modul learn modul schedul knowledg agent network resourc inform knowledg organiz organiz design learn compon agent belief db gpgp coordin modul commit db exchang inform task structur client goal criteria exchang inform exchang inform detect system probe network inform schedul data flow except notif data compon state inform updat updat user exchang domain inform updat updat produc reschedul request updat reschedul request updat updat updat notif figur prototyp agent architectur problem solver problem domain solv option system constraint domain plan resourc constraint client goal criteria modul local control commit agent commit agent commit import criteria ramif local control context satisfi commit violat commit schedul select execut schedul coordin modul specif local problem solv action suit current local context modul context schedul select execut envelop defin rang accept action outcom data flow figur primari agent control compon resourc constraint describ intim detail primit action primit action call statist discret probabl distribut cost durat qualiti abstract concept describ contribut action problem solv applic notion correspond model qualiti durat describ amount time action model method execut cost describ opportun cost inher perform action uncertainti dimens implicit perform agent reason certainti action durat uncertainti appli task interact facilit hinder time facilit qualiti time qualiti quantif method interact taem regard perfect scienc task structur programm problem solver gener estim perform characterist primit action estim refin time learn reason typic replan reschedul unexpect event occur ground figur simplifi task structur emit big inform gather describ portion inform gather process task construct product model retail pc system term expect durat enabl arc gather task model fact review gather method product order gather review task interact model taem bound disabl task interact interest coordin research identifi instanc task assign agent interdepend implicit joint goal joint problem solv activ coordin motiv exist interact return join facilit hinder task interact model soft relationship result produc task benefici harm task case exist activ gener qualiti recipi task reduc cost durat monei resourc build pc product object basic product inform queri extract vendor queri extract maker gather review search process zdnet review search process pc world queri process consum report queri extract pc connect extract pc mall consum limit multipli multipli multipli task method resourc task subtask relat kei enabl figur simplifi subset inform gather task structur function defin perform subtask relat perform parent task method emploi achiev true seq subtask result qualiti sum determin qualiti parent altern wai achiev goal task order recent addit taem denot denot denot number children denot qualiti item qualiti task qualiti ith child qualiti task method perform sampl defin taem ffl subtask perform minu order ffl sum subtask perform order ffl nq subtask perform order task initi failur perform child min result qualiti parent task ffl nq number subtask perform gener task select ffl nq subtask perform ffl nq subtask perform order ffl seq seq seq seq prefix case denot sequenc prefer subtask suffix denot function perform result seq sum taem support model task arriv point individu deadlin earliest start time task belong schedul taem task structur process develop taem constant tension power inher work structur result model schedul optim lend flexibl approxim process strategi local agent control schedul agent local expert make control schedul role domain action domain problem solver action local agent goal criteria prefer type local agent resourc constraint express gpgp coordin modul gener idea evalu option light constraint prefer sourc find achiev select task address schedul problem frame term taem task structur emit domain problem solver schedul problem solv activ model taem languag major find set action achiev sequenc find sequenc action soft produc schedul meet dynam goal certainti client taem model multipl approach achiev task durat characterist primit specif enabl taem client reason action taem task multipl approach achiev task approach certainti characterist contrast classic schedul taem schedul object sequenc set unord action find sequenc set action suit client certainti examin current current option decid action target contextu make schedul requir sophist heurist approach schedul task inher comput complex exhaust search techniqu find optim schedul deadlin resourc constraint exist complex task prevent singl heurist produc optim schedul cope satisf respect goal criteria respect search solut space satisf translat techniqu reduc search space make schedul problem process includ gener partial approxim guid schedul goal design criteria desir approxim schedul call provid overview schedul solut space heurist make set action perform order consid cope complex group heurist action order task resourc commit made problem solv constraint heurist algorithm reduc action order problem polynomi level worst case heurist error correct secondari set improv heurist act safeti net catch potenti correct error produc heurist approach cope comput complex client goal criteria focu reason schedul approxim complet schedul approach methodolog effect aspect schedul problem soft amen satisf approach task interact commit made agent import relax gpgp manag interact gpgp partial global agent tool interact agent coordin joint activ gpgp domain approach coordin coordin modul local control post constraint agent local schedul gpgp coordin modul gener commun commun agent local commun make break task relat commit agent coordin modul modular coordin subset appli coordin depend coordin desir gpgp defin coordin mechan formal detail share view basic coordin mechan handl exchang local view agent detect task interact exchang local view agent detect coordin task interact mechan exchang exchang exchang inform exchang part local view exchang entir local view commun coordin mechan coordin mechan local view task understand exist task interact coordin commun result coordin mechan handl commun result method execut agent govern minim polici result satisfi extern commit polici minim result final result task group polici result commun mechan meaningless mechan mechan form commit avoid redund mechan deal detect redund pick agent random execut redund method question agent commit perform action agent commit denot agent carri task predetermin time type redund question simpl duplic contrast redund gener similar result method handl hard task relationship enabl pictur figur denot hard task relationship coordin mechan deal task interact commit predecessor enabl perform task deadlin handl soft task relationship soft task unlik hard interact option coordin mechan attempt form commit predecessor soft interact perform method question method receiv end interact mention gpgp coordin modul modul local control place call local schedul commit repres deal gpgp made perform method time deal gpgp make agent commit fall deadlin commit type commit denot agreement execut method time agent result method execut perform agent form deadlin agent plan activ base expect receiv result deadlin earliest start time commit commit denot agreement start execut method prior time type commit deadlin commit agent scenario commit denot agent execut time start execut time commit commit weak simpli denot commit execut method time dont commit commit denot agreement perform method interv coordin share resourc salient featur coordin includ domain independ approach exchang inform construct partial global worth driven view task action inform exchang polici coordin subset mechan independ depend current context loom coordin agent reli underli oracl analysi procedur determin bind variabl exchang agent dialogu agent good idea result provid agent order commit effect world inform gener provid schedul gpgp requir analysi code detect task interact determin inform polici gpgp mechan embodi analysi aspect coordin problem coordin protocol aspect problem integr concern make extend protocol difficult built code isol world address problem separ analysi procedur specif agent coordin protocol label current gener agent coordin tool primarili histor reason goal project develop approach coordin mechan separ coordin protocol support analysi code coordin protocol easili modifi adapt context step verif tool reimplement function includ fairli simpl coordin protocol coordin natur main object work territori cover gpgp gpgp group analysi function protocol specif singl bodi approach coordin protocol finit state machin model state denot state transit call commun action analysi code approach specif akin cool work differ interact underli agent control machineri fsm script process fsm interpret interpret emit java code incorpor coordin bean integr gener java agent framework coordin bean interact rest agent compon mechan direct invoc support featur framework featur fsm model interpret ffl fsm variabl enabl protocol store inform explicitli addit implicit inform state store commit time agent ffl share fsm variabl enabl interact set interrel task sequenti point synchron serial effort synchron phase entail share variabl pass bind inform pass coordin bean standard agent data structur knowledg intuit effici sort oper insid coordin machineri gener agent control structur design embodi issu handl interact approach unclear interact ffl timer enabl machin set timer wait event occur timer enabl produc window timeout durat specif global default ffl event registr creation event gener fsm agent class object agent framework term event gener event catch interact explicit priori knowledg interact ffl part event fsm initi fsm detect dialogu fsm handl dialogu ffl inherit coordin protocol subclass special facilit rapid protocol develop simpl special exist protocol ffl pre post condit transit transit set action test fsm place transit set action place transit complet ffl except fsm throw catch except fsm handl timeout condit event drastic chang execut flow succinct fashion altern complet specif except transit state messag receiv messag code execut messag process code execut messag process inform fsm environ watch condit test fsm variabl declar fsm variabl notifi ack comput commit time commit reject commit evalu commit set counter comput commit time commit reject counter branch reject branch accept branch accept add firm local commit accept add firm local commit notifi ack timeout resend track resend give number regist condit figur initi fsm coordin hard task interact messag receiv messag code execut messag process code execut messag process inform fsm environ watch condit test fsm variabl declar fsm variabl notif decid coordin set variabl ack commit evalu commit set counter comput commit time commit reject counter branch reject branch accept branch accept add firm local commit accept add firm local commit branch ack dont branch figur respond fsm coordin hard task interact figur show exampl fsm handl coordin hard task interact tempor sequenc task fsm figur design handl format singl commit outstand research question determin agent model task interact coordin case interact task ff fi fl chain interact requir coordin chain multipl independ multipl interact share variabl case agent agent multipl task interact current handl multipl concurr asynchron agent handl singl dealt multipl task interact interact fsm issu interdepend task coordin interdepend interdepend task interdepend task stem assign task agent interact revisit issu interact potenti larger section identifi issu interact interact machineri agent control machineri agent dialogu coordin mechan potenti oper held perform gpgp style coordin gpgp deal tempor sequenc task explor task constraint assign set agent style coordin feasibl analysi solut enact base assumpt task gener dynam problem solv agent problem solver extern client gpgp assum process determin set candid task schedul coordin note taem model altern wai perform gpgp problem simpli coordin schedul set primit action perform action perform base util feasibl gpgp choic limit set task action emit problem solver gpgp pertain detect sequenc activ resolv share pertain determin goal agent current work integr requir address issu task alloc agent determin task perform global perspect note issu task alloc problem assign member set individu agent belong set candid agent requir knowledg capabl resourc agent knowledg structur task view interdepend determin task agent network pursu matter process gener activ requir agent engag requir coordin convei inform capabl agent inform pertain state problem solv network gener respond reflect version reason singl fsm adapt interpret output version remov analyz fsm reachabl relat issu top level bottom level coalit format dialogu matrix organ joint goal remov joint goal joint goal agent attempt achiev task view commit accept commit result result result influenc constraint feedback influenc constraint feedback influenc constraint influenc constraint feedback gpgp result share gpgp coordin interact detect commit format bdi share plan task goal select task assign evalu agent capabl organiz structur figur level interact pertain concern oper level detail interact product assign member feasibl case face interact held determin held determin task held determin task feasibl task perform respect level simultan figur illustr idea made upper level set polici lower level decid pursu task ff fi upper level determin held end feedback process explor feasibl task select upper level situat set task select agent attempt perform requir action discov agent network lack suffici resourc carri activ address problem task interact tempor constraint make difficult ascertain process attempt coordin schedul choic task pursu network modifi return issu interact interact explicitli model handl requir negoti style interfac machineri level begin agent level obviou work organ comput organ system entail place level abstract agent determin structur problem solv place level interact explor integr base coordin bdi approach agent control bdi framework upper level agent control determin task perform perspect coordin schedul activ carri tool lower vice versa machineri hold stratif move food chain examin level gpgp agent exchang local inform construct partial global view rest world agent carri dialogu attempt handl task interact activ fall gener umbrella feasibl solut enact act commun result view type activ machineri commun result carri activ inher case machineri possibl interact level result commun affect exist result agent engag agent problem solv state evolv attempt identifi issu interact agent provid reason interact research question worth address agent commun coordin commun benefit integr technolog integr technolog lead issu interact level control level base work level scale system open environ main issu formal explicitli repres reason interact requir step notion level interact issu uncertainti uncertainti agent coordin taem explicitli reason certainti action begun reason role uncertainti coordin intuit uncertainti question ubiquit appli level agent refer martin andersson sandholm level commit contract myopic strateg agent fifteenth nation confer artifici page barbuceanu mark fox languag describ coordin multi agent syst ms intern confer system page ana lc victor ping xuan adapt organ design domainindepend diagnosi comput scienc technic report massachusett februari norman carver victor lesser testb research distribut situat extens model extern evid intern confer multiag pr cohen hj levesqu commun action artifici agent intern confer system page dean boddi analysi plan seventh nation confer artifici page st august sycara intellig adapt inform agent journal intellig inform keith decker li coordin hospit patient schedul intern confer system page keith decker environ center analysi design coordin mechan phd keith decker task environ center simul simul comput model institut group aaai keith decker victor lesser coordin assist mix human comput agent system concurr engin page concurr technolog corp umass cs mark david michael luck interact protocol agenti intern confer system page edmund thoma montgomeri coordin distribut search hierarch behavior space ieee transact alan keith victor lesser interfac schedul aaai workshop model conflict umass cs bryan horl reusabl compon architectur agent construct umass depart comput scienc technic report octob bryan victor regi ana ping xuan diagnosi integr part adapt comput scienc technic report massachusett januari eric horvitz reason vari uncertain resourc constraint seventh nation confer artifici august nr jen control cooper problem solv industri system artifici david michael regi victor lesser learn quantit knowledg multiag coordin umass depart comput scienc technic report stanlei sutton jr leon design process languag acm sigsoft symposium foundat softwar page septemb gal kaminka milind tamb wrong improv robust social diagnosi fifteenth nation confer artifici juli toru coordin protocol descript system intern confer system page yanni labrou tim finin kqml specif comput scienc technic report maryland baltimor februari victor michael bryan brett anita regi thoma ping xq zhang system intellig environ control intern confer autonom agent victor keith norman alan daniel nagendra thoma wagner evolut gpgp coordin framework comput scienc technic report massachusett januari victor bryan frank anita thoma xq zhang inform gather agent fifteenth nation confer artifici intellig juli umass cs technic report nagendra lesser cooper inform distribut problem solv approach comput scienc technic report journal softwar special issu develop agent base sandip sen effect misconcept reciproc agent intern confer autonom agent page munindar singh develop formal specif coordin heterogen autonom agent intern confer system page ira phillip jefferi mark design polici joint intent theori intern confer system sugawara victor lesser learn improv coordin action cooper distribut environ machin milind ali gal ion talli explicit model teamwork robot world cup competit tamb flexibl teamwork journal artifici intellig regi bryan thoma victor lesser surviv simul multiag adapt coordin intern confer model umass cs thoma alan victor lesser heurist task schedul intern journal approxim special issu version umass cs thoma wagner victor lesser ubiquit satisf agent control aaai symposium satisf thoma victor brett anita ping xq zhang improv divid conquer work ping xuan victor lesser handl uncertainti commit umass depart comput scienc technic report zilberstein stuart russel effici reason intern confer ai plan colleg june deal schedul reschedul iter repair zweben intellig chapter morgan 
Agents/1617 1 istituto la ricerca povo itali model check multiag system serafini august technic report fl istituto di limit distribut notic report submit public itc copyright accept public issu technic report earli dissemin content view copyright distribut itc prior public limit peer commun specif request request fill reprint legal obtain copi articl model check multiag system giunchiglia serafini universita degli studi di itali irst istituto di itali abstract model check success techniqu appli design verif finit state concurr reactiv process paper show techniqu lift applic multiag system approach technolog tool develop model design verifi multiag system modular increment effici model check algorithm introduct model check success automat techniqu design verif finit state reactiv sequenti circuit commun safeti critic control system evid model success approach formal method verif order induct theorem tableau base reason modal nowadai effici develop tool implement model spin tool successfulli appli real case industri goal paper show model check applic multiag system variat technolog tool develop model modular design verif multiag system featur exploit huge amount technolog tool develop model check featur order deal real world complex system discuss model check model concurr reactiv finit state process model agent concurr reactiv finit state process paper accept public journal logic comput publish time materi paper definit develop version logic model check algorithm properti call bdi desir intent detail bdi specif agent orthogon tempor aspect aspect kei idea underli approach aspect separ practic thing work ffl tempor evolut agent treat bdi atom atom formula express atom proposit fact formula talk bdi attitud ffl deal bdi attitud fact agent bdi attitud agent model fact access process bdi time verifi truth bdi atom af simpli test af oe hold bdi attitud essenti control process oper iter obviou case nest bdi attitud agent model set process achiev requir set section techniqu develop model check achiev modular process bdi attitud enforc relat independ defin paper structur section describ motiv formal studi rest paper basic ingredi model check proposit tempor logic write languag describ system set verifi finit state model check procedur effici automat determin specif satisfi graph gener system automaton section describ ingredi multiag model check descript increment standard model check notion adopt ctl proposit tempor logic state specif import section show algorithm effici visit smallest conclud detail discuss relev work multiag system conclud remark motiv scenario involv receiv sender continu read subject sensor standard read inform believ correct knowledg subject order minim traffic receiv acknowledg fact back initi state proposit loop initi state loop figur rs algorithm implement scenario fipa compliant architectur network interact figur give algorithm descript commun languag subject inform exchang truth proposit atom return messag sender receiv content fipa primit put commun messag read read standard input oper repres belief perceiv dualli notic commun protocol belief behav coincid figur model fact commun protocol behav correctli nest belief oper belief principl true figur intuit mean necessarili reach state perform action algorithm figur codifi fact protocol implement inform flow fact deliv messag deliv properti prove input languag spin model checker initi state proposit loop set proposit begin end begin end end set proposit begin end begin end figur commun protocol algorithm ffl bdi eventu ffl live read futur notic present aspect interest formal involv multiag formal commun agent agent bound bdi attitud level nest kind properti system properti properti agent multiag tempor logic intuit logic call tempor basic composit formal tempor formal bdi attitud exploit fact introduc step start tempor compon ctl give detail present hierarch metalog formal bdi attitud variat logic introduc integr logic ctl ctl branch time proposit tempor logic turn languag semant languag set proposit set ctl formula oe defin induct oe oe intuit oe mean path oe true mean true state futur oe true state mean exist path true state futur oe true state abbrevi def def oe oe def af oe def ef oe def ag oe def oe def ax oe def semant semant ctl formula standard tempor semant base ctl structur tupl set initi total binari relat label associ state set proposit atom true path infinit sequenc state rs satisfi formula oe ctl structur state defin ffl iff ffl iff ffl oe iff oe ffl oe iff sr ffl iff path ffl iff path oe ffl oe iff oe ds bs br dr ir figur set view section turn semant model situat set agent agent intent agent adopt ffl oe mean agent believ ffl oe mean agent desir ffl oe mean agent intent bring call bdi oper agent simpli bdi denot bdi oper agent ig set bdi attitud oi set set finit string form call ff oi view view oi repres nest bdi attitud empti ffl intuit ffl repres view extern observ behavior multiag system depend extern observ repres person design select process multiag system privileg statu section extern observ point view commun protocol view ffl belief commun view br belief agent final belief belief formal view view view principl view instanc view ir rs view arbitrarili deep nest bdi attitud figur graphic repres situat dark circl repres view exist white repres notic principl infinit case real applic hard applic deeper depth nest bdi make reason multiag system feasibl fact gener complex grow combinatori depth bdi attitud import notic figur crucial notion view tree root view extern observ bs notic view agent agent gener agent section model fact belief associ logic languag ff view ff oi ff languag express true ff languag ffl speak multiag system formula ffl ffl mean true believ languag ii languag adopt repres desir languag belief js section formula dr mean receiv desir true sender believ notic put restrict languag ff oe atom formula ff oe formula studi condit modifi order captur interest empti languag ffl empti talk multiag system extrem flexibl modular defin languag view essenti independ import featur system crucial featur scale complex system agent view agent semant defin semant famili languag fl ff drop index ff oi understand semant understand kei fact hand semant formula depend view formula view written bs express fact believ true formula view br express fact believ true semant associ local view ff set ff interpret ff hand formula view intend mean instanc believ believ true impli subset interpret view truth valu formula intend mean captur notion compat introduc notion tree defin tree famili set ft ff ff ff interpret ff defin compat relat set tree ts element ff fi fi assign truth valu formula intend mean detail discuss intuit technic interpret ffl interpret bs interpret br interpret ffl br bs figur tree section figur show tree associ ffl interpret white box number associ bs associ empti set interpret view relev defin semant start satisfi local view view ff satisfi relat ff formula ff denot satisfi relat symbol context make clear relat step defin satisfi account tree notat denot satisfi tree extend satisfi relat set model ff oe iff ff oe tree satisfi formula formula satisfi element tree associ view introduc notat mean ff ff ff readi defin notion model hm satisfi hm structur formula view definit set tree fl ff close hierarch fl ff ffl ff ff ff impli ff oe intuit underli definit detail condit extern observ consist view condit view correct bdi time oe hold view oe hold view level condit dual properti ensur complet view hm structur formula oe view ff oe read oe true oe hold oe satisfi view defin ff oe iff ff oe intuit order check satisfi oe view ff check interpret ff compat tree result show hm structur defin logic essenti modal oper multimod logic number agent turn semant defin kind languag ff ctl languag view languag fp ff famili set proposit atom ff definit languag call languag fp ff view bdi extend proposit atom bdi atom famili languag fp ff famili ctl languag fl ff ff smallest ctl languag set proposit atom ff bdi atom oe formula oe model fact set rg extern point view commun protocol set proposit atom commun protocol proposit atom worri rs bdi ffl sender belief proposit atom action read send messag receiv messag pb proposit atom languag true state action result easili extend result stronger modal paper stick modal perform effect action sender intern state simpli proposit atom true analog intuit proposit atom form effect action set variabl true sender ascrib receiv belief proposit atom fpg receiv belief proposit atom action send messag receiv messag bdi attitud ff ff famili set proposit atom set bdi atom set defin famili languag fl ff ff ff empti languag formula form oe lb bdi atom form oe oe final ffl bdi atom form oe oe form oe oe lb formula ffl mean receiv futur messag semant semant famili languag defin term hm semant local view set ctl structur ctl structur li state ctl structur obtain replac initi state definit structur fp ff hm structur famili languag fp ff ctl structur ff state ff structur kind hm structur notion satisfi structur formula oe view ff oe read oe true view defin ff oe iff ff oe definit understood basi crucial mutual nest ctl oper bdi oper view structur ctl structur differ initi state redund admit sake clean formal semant section show structur present structur nest ctl oper insid bdi oper ff oe comput notion satisfi ctl structur tree link fact bdi atom hold initi state ctl structur view fact argument hold initi state ctl structur view instanc tree ctl structur view ffl fm hs ig bs fm hs ig br fm hs ig fm hs ig model fact process commun protocol state state protocol believ sender process state sender state believ receiv process state protocol believ receiv process state exp intuit mean sender state possibl mean formula exp bs compat formula true exp true state ctl structur figur exampl constraint kind tree crucial observ concern nest bdi oper insid tempor oper oper involv bdi atom treat ctl jump instanc formula assess truth assess truth futur state ctl structur li establish request tree access view ctl structur fact tree connect ctl structur hold initi solut request initi state ctl structur ffl fact tempor oper state fact state ctl oper repeat state definit bdi live properti introduc section formul bs af ffl oe afb afb turn properti hold intuit hold sender read true bs br exp protocol receiv sender exp exp sender receiv figur tree structur section inform receipt acknowledg set true concern proof statement suffici notic fact properti proof system proof carri notic initi set depend truth eventu proof statement prove step step finit state present system section multiag finit state machin structur gener infinit structur ctl structur infinit number label function map infinit number hm structur infinit number tree infinit compat tree infinit branch infinit number attitud agent infinit number infinit long tree case unbound nest bdi interest extend model check multiag model check model check deal finit ctl ctl structur finit set label function map finit number atom crucial observ finit ctl structur finit state machin fsm object li solut extend notion fsm multiag finit state machin roughli finit set fsm step direct restrict finit hm hm structur finit number finit number view ff ff empti limit number depth finit hm structur infinit number bdi atom finit number process model fsm solv problem introduc notion explicit bdi atom formal fl ff famili finit subset bdi atom ff element call explicit bdi atom definit fl ff famili languag fp ff multiag finit state machin ff ff fl ff total function ffl view ff oi ae oi oi associ finit set ff fsm languag ff view ff oi oi ff condit condit ffl section condit definit hm structur section reason deal finit deal finit set atom construct section essenti thing work concentr process view proposit atom proposit atom specif bdi atom defin oe bdi atom ff oe formula set explicit bdi atom bdi atom set specif intuit explicit bdi atom chang implicit bdi atom chang consequ explicit bdi atom identifi set state variabl space state construct directli transit finit state automaton fix set state fsm extract specif differ initi state state variabl initi state extract top part specif view fsm initi fsm choic initi state initi state complet fsm multipl fsm view defin specif languag issu detail fsm process leav reader definit state transit fsm view br initi state complet defin fsm view proposit atom view concentr explicit bdi atom fsm ffl fsm gener protocol fsm fb fsm gener sender fb set explicit bdi atom empti fsm fsm gener receiv empti set fsm section give algorithm specif view true inform natur languag specif state variabl fsm state variabl formal fact initi state state transit fsm view ff bdi atom specif mention notion step give notion satisfi start notion satisfi ctl formula fsm state notion defin ctl structur determin satisfi proposit explicit bdi atom formula belong formula machineri bdi attitud machineri order deal number bdi atom memor set implicit bdi atom view written defin subset bdi atom ff explicit bdi fo oe ff defin foe oe consist formula oe explicit bdi atom oe true restrict explicit bdi set formula satisfi condit correct definit hm definit section defin satisfi suffici fact comput suffici ctl satisfi compar result relev ctl exploit comput implicit bdi atom satisfi condit correct complet definit ff li ff state formula oe ff satisfi relat oe defin proposit atom explicit bdi fsm satisfi proposit connect ctl fsm oe implicit bdi iff state oe oe oe iff ff oe iff ff ff oe iff ff oe definit item crucial step conjunct element set formula order comput formula oe oe implicit bdi atom end section bdi oper strength modal oe oe theorem view oe oe theorem view set fo oe oe remain item natur counterpart respect definit structur item state fsm satisfi formula formula satisfi initi state item state formula satisfi view satisfi fsm view final item state label formula ff oe satisfi oe satisfi view label prove defin satisfi formula ffl oe afb afb notic formula implicit bdi afb reason fsm ffl item definit impli item definit impli fb state oe fb fsm gener sender sender send sender send messag oe oe oe af oe valid ctl oe afb item definit afb oe oe af oe valid ctl afb afb case item sender send messag oe item definit infer ffl fsm gener protocol ffl path start state item definit impli state oe fsm gener oe item definit impli fsm ffl gener path start state item definit impli fb state oe fb fsm gener oe afb item definit afb notic path start state path start state afb conclud path start afb definit ctl afb afb reach case item item conclud oe afb afb ffl ffl oe afb afb step relat notion satisfi notion satisfi structur structur view satisfi set formula proposit structur finit hm structur vice versa proof view ff set ctl structur finit occur ff tree label function ctl structur label infinit set bdi atom restrict finit subset bdi atom explicit bdi ff ctl structur select subset set ctl structur defin fo oe oe oe oe formula ctl structur view start build label dag acycl set node fs label dag disjoint union fs ff ff li ff sg set label arc defin arc label connect fsm fs ff fs iff foe oe structur tree subgraph label graph defin set formula satisfi ff comput extend label function fsm satisfi implicit bdi atom model check introduc data structur algorithm data structur basic oper standard ctl model check algorithm extend label function fsm consid proposit subformula formula model check call extend fsm fsm context make clear result oper gener extend fsm reli fact label function explicitli defin truth atom problem fsm label function defin implicit bdi truth left truth valu implicit bdi atom occur formula model check definit satisfi definit fix problem implicit bdi state oe crucial observ gener label function finit properti independ formula model store data inform bdi oper coi call compat relat relat defin subset explicit bdi atom view ff coi state foe oe start view ff set explicit bdi atom coi collect fsm state view satisfi argument chosen explicit bdi atom notic version compat relat defin section hm restrict explicit bdi atom subset priori explicit bdi atom relev comput truth implicit bdi atom implicit bdi atom evalu state easili hf coi iff oe iff hf coi oe oe implicit bdi atom set explicit bdi atom satisfi state ff model check algorithm reli global data data structur set fsm ff data structur oper set explicit bdi atom data structur modal oper compat relat coi algorithm multiag model check algorithm view ff formula oe ff model check return true ff notic model check view subpart multiag ff ffl model check multiag ff length model check singl ff length model check view agent algorithm shown figur call algorithm view ff set formula input view ff set formula ff label subformula formula return truth simpli test oe label set initi state fsm ff section fsm satisfi formula initi state satisfi denot set subformula oe oe atom subformula insid distinguish main phase global variabl fo oeg oe explicit bdi atom ff ff ff set fsm ff coi ig fsm state algorithm ff oe end end algorithm oe oi foe oe endif end ff implicit bdi oi foe oe hf coi hs end end end endif end end figur multiag model check algorithm phase initi line collect subformula formula phase model check implicit bdi atom phase correspond loop loop consid turn pair oi step comput set set formula oe argument implicit bdi atom oe subformula notic step perform set explicit bdi atom set implicit bdi atom set infinit notic knowledg formula model check restrict finit set implicit bdi atom relev satisfi formula step call view set phase visit tree structur model extend label function visit fsm leav tree view model check implicit bdi due fact implicit bdi atom occur set formula input notic tree visit tree constitu view comput implicit bdi atom mention top level goal formula input phase phase loop fsm current view ff loop iter perform phase label implicit bdi atom phase correspond level loop enter view implicit bdi atom occur input formula algorithm extend label function true implicit bdi atom oe occur step comput definit satisfi implicit bdi atom set true explicit bdi atom state fsm view ff hf fsm state view compat true explicit bdi atom comput innermost loop comput store argument implicit bdi atom occur true pair hf end loop set implicit bdi atom true current state set extend label phase model check point state current label atom proposit explicit implicit bdi occur suffici appli usual ctl model check algorithm input fsm set formula extend label function subformula instanc notic call model checker black box result state solv model check problem theorem li ff ff state oe iff appli construct oe state proof proce induct nest bdi oper occur oe oe ctl formula subformula oe bdi atom phase leav unchang fact call iter loop phase perform correct consequ correct assum thesi hold maximum number nest bdi oper bdi atom equat state iff hf coi induct phase fsm hs view phase algorithm extend label set state ff bdi atom pair hf coi iff phase label function extend atom phase appli ctl model extend label function subformula oe thesi correct ctl model check algorithm relat work work paper root previou work develop author collabor idea defin modal logic proposit attitud descend tree hierarch metalog introduc idea exploit modular exist system order model complex multiag system introduc idea give semant hierarch metalog chain introduc idea treat belief atom proposit atom exploit fact order achiev comput effici introduc work paper complet deal problem tempor show implement effici multiag model check literatur multiag system theori huge start semin work bratman lot work develop bdi architectur relev work line found formal verif multiag system notic except wooldridg fisher work concurr approach issu proof theoret point rao georgeff work develop satisfi base procedur class bdi logic closest work work rao georgeff face problem appli model check techniqu multiag system similarli emploi class logic obtain combin tempor logic logic bdi attitud result logic similar main differ multiag system uniqu languag semant consist uniqu kripk structur tempor access relat access relat bdi attitud singl make lose modular increment structur correspond agent specif structur model notic import featur give import advantag definit model check algorithm deal natur case bound implement multiag model check directli standard model check implement algorithm visit smallest properti possess algorithm label world improv work dont face problem automat gener model start specif multiag system notic current standard model check languag extend gener model multiag system paper defin base procedur multiag system shown compon model check specif languag finit state model check algorithm extend multiag system approach achiev technolog tool develop model multiag system procedur defin effici begin lot work remain individu class interest defin interest subclass logic defin implement logic full system construct extens model checker test result system applic interest refer bratman plan practic reason harvard long model check intern summer school deduct program fipa foundat intellig physic agent fipa draft giunchiglia giunchiglia ideal real belief belief practic intern confer formal appli practic number lectur note artifici page springer giunchiglia local model contextu reason local compat technic report februari aaai fall symposium context kr nl giunchiglia sebastiani build procedur modal logic proposit procedur case studi modal proc confer autom lectur note artifici intellig august springer verlag report report giunchiglia serafini hierarch logic modal artifici report itali editor intern confer springer gj design valid comput protocol prentic jen control cooper problem solv industri system joint intent artifici kl mcmillan symbol model check kluwer academ rao georgeff model ration agent bdi architectur intern confer principl knowledg reason morgan kaufmann rao georgeff approach verif situat reason system thirteenth intern joint confer artifici intellig page rao georgeff formal model procedur multiag system technic note australian ai level la june wooldridg fisher procedur tempor belief logic gabbai tempor logic intern confer volum page juli 
ML/1161 6 learn detect salient object natur scene visual attent laurent koch california institut comput neural system program mail code ca abstract primat visual select attent rapidli select conspicu imag locat analyz detail select guid featur extract detect candid salient locat base local properti featur type orient difficulti inform modal combin singl control visual attent quantit compar featur combin strategi learn linear natur color imag introduct visual attent primat real conspicu target clutter visual environ reproduc target detect capabl artifici system import navig aid robot navig base studi human studi believ visual work support nation scienc foundat offic naval research special werner ritter provid traffic sign imag pietro perona comment discuss attent act akin neuron visual world enhanc restrict area attent progress cortic hierarchi highlevel pattern recognit locat spotlight control featur extract crude segment conspicu locat entir visual massiv parallel compet attent attent sequenti salient imag locat analyz detail koch visual attent seemingli perform break complex scene understand sequenc circumscrib pattern recognit problem volit compon attent form attent studi disregard compon studi model function account visual attent primat baluja model share similar gener architectur topograph detect local spatial discontinu orient attend locat attend locat inhibit featur map intens orient color linear filter differ normal featur combin maximum detector salienc map input imag weight coeffici binari target mask learn fig gener architectur visual attent system studi earli visual featur extract parallel entir visual scene featur extract achiev linear filter featur type color oper extract local spatial discontinu featur type featur map combin uniqu studi inform combin modal import color discontinu compar orient involv learn target region target combin maximum detector select salient locat salienc map shift attent locat subsequ suppress system focu salient locat flow achiev mechan akin biolog visual recept field recept field properti approxim filter gabor filter orient featur map visual modal combin uniqu map koch salienc map sequenti order focu attent central studi biolog artifici combin featur visual modal unrel dynam rang color uniqu salienc map model assum simpl summat featur linear combin weight quantit compar combin strategi natur color simpl summat scale fix dynam linear combin weight imag addit combin enhanc featur map isol peak suppress featur map uniform activ model detail model present briefli schemat fig import rememb type color orient extract separ combin uniqu activ control attent fusion inform difficulti combin featur map uniqu scalar salienc map repres priori compar dynam rang extract mechan larg number featur map combin color orient salient object strongli map risk mask salient object present larger number map simpl approach solv problem normal featur map dynam rang sum featur map salienc map priori weight featur refer strategi intens contrast color contrast orient contrast ms ms ms ms attend locat input imag salienc map fig oper model natur imag featur combin strategi exampl found learn learn introduc specif target detect featur map global multipli weight factor final input salienc map sum featur map featur map weight train base featur map insid manual outlin imag region desir target learn procedur weight featur map consist comput global maximum minimum map maximum insid manual outlin target maximum target updat weight addit learn rule independ map dynam determin learn speed posit weight learn procedur particip salienc map featur map show higher peak activ insid target map remain system weight converg comput initi salienc map attent scale fix rel weight featur map potenti diverg addit learn rule avoid constrain weight fix sum note local maxima activ imag averag activ area local activ import visual extend region small strong peak peak highli salient immedi averag activ extend region low amplif simpl normal consist global promot featur map small number strong peak activ man global suppress featur map elicit compar peak numer locat visual scene normal denot consist normal featur map dynam order elimin amplitud differ due dissimilar featur extract find global maximum averag local global multipli map local maxima activ consid compar map ignor homogen area compar maximum activ entir map averag activ spot measur activ locat averag differ activ locat stand strongli promot map differ map uniqu suppress normal replic biolog inhibit neighbor similar featur inhibit cannon normal oper close implement oper result discuss natur color imag evalu featur combin strategi consist imag red target demonstr simplest form featur map system specif encod main featur target explicitli detect system featur map consist imag vehicl emerg triangl target complic form special target uniqu imag conjunct red color ffi ffi ffi orient featur type repres system separ independ featur map consist imag acquir video camera mount passeng side vehicl driven german traffic sign traffic traffic sign target outlin binari target mask creat target consid detect focu attent intersect target imag disk radiu pixel complet coverag entir imag requir locat system perform random visit averag locat find small target imag imag split train set imag traffic test set imag learn train random pass set halv learn speed pass compar result obtain test imag set featur combin naiv model dedic normal featur weight set model dedic train dedic normal featur weight learn train set retain test set challeng target mediat detect version model imag version model immedi find target discard ensur perform result summar tabl tabl averag number detect standard red test set emerg triangl test set traffic sign test set imag naiv train red triangl traffic traffic sign found sign found naiv repres simplest solut problem combin featur map uniqu salienc perform simpl normal prove effici elimin featur map numer peak activ intens map imag larg variat illumin addit learn rule prove effici special gener model limit special obtain global weight featur learn enhanc map suppress poor gener expect learn larg varieti object type linear train emploi sum featur learn conjunct model train emerg triangl attend strong obliqu edg red color present red blob absenc obliqu orient extent human train number attend locat number target imag fig number attend locat standard find traffic sign real color normal featur compar optim perform detect learn conjunct featur remain controversi remark train model perform altern model studi traffic wide varieti shape color textur sign result entir traffic sign final present fig ideal traffic sign detector visit locat target fig perform system close dramat chanc perform good obtain train remark fact oper specif traffic sign oper provid reliabl detect salient imag repres approxim human salienc altern studi compar simpl strategi combin multipl featur map visual modal singl salienc map introduct simpl learn scheme prove effici detect specif broad special gener model remark good perform obtain normal replic biolog spatial competit salienc addit learn normal strategi provid perform improv model summat weight summat featur combin strategi refer baluja baluja da select attent visual monitor control robot robot autonom cannon mw cannon sc model inhibitori interact effect perceiv vision vol pp jp visual salienc monkei tk tg koch sp vlsi circuit visual neural inform process system mi pp koch koch select visual underli neural human koch model visual attent rapid scene ieee transact pattern analysi machin intellig gil mechan dynam static scene optic architectur attent ba ch anderson ch dc van neurobiolog model visual attent invari pattern recognit base dynam rout attent theori cognit jk sm davi visual attent select artifici jm search model visual bulletin 
ML/245 6 transact vol june make popul inform evolutionari artifici neural network xin senior yong student ieee paper concern simultan evolut artifici neural network architectur weight current practic evolv ann gener final result paper approach form final result combin individu gener order make inform popul approach popul ensembl combin method integr work integr ann modul evolutionari learn make popul inform linear combin method investig paper illustr idea data set show algorithm produc integr system outperform individu result confirm popul inform singl individu evolutionari learn exploit inform improv gener learn system index evolutionari artifici neural evolutionari modul learn introduct evolutionari artifici neural network refer special class artifici neural network evolut fundament form adapt addit learn evolut introduc level evolv learn paramet rule studi wide recent year provid automat method design approach studi evolut learn framework paper concern evolut architectur weight evolutionari algorithm evolv architectur weight studi evolv effect effici issu gener final result evolutionari process overlook final result individu gener gener attempt made exploit popul inform practic treat evolutionari learn evolutionari manuscript receiv februari work support part australian research council small grant scheme author comput intellig school comput south australian defenc forc act australia publish item identifi ieee practic argu paper differ learn optim exploit evolutionari comput learn optim paper show popul inform singl individu popul inform improv gener learn system paper linearli combin individu gener form final integr system idea combin modul studi ann field statist attempt made evolutionari learn popul inform form final system carri studi problem demonstr effect approach automat ann design tool base evolutionari program evolv popul perform individu evolv compar integr system linearli combin individu gener integr linear combin weight found algorithm pp outperform individu term gener problem test reason linear combin paper simplic find combin show import popul inform advantag combin improv result expect nonlinear combin method rest paper organ section ii discuss differ evolutionari learn optim opportun exploit differ learn section iii describ evolutionari ann design individu gener final output result real world problem present section section iv combin method studi find combin demonstr effect approach simpl method section conclud summari paper futur research direct ii evolutionari learn optim learn formul optim problem machin learn field backpropag train feedforward train ieee transact vol june process call learn process bp wide learn algorithm bp essenc optim algorithm minim error function squar learn problem typic optim problem numer analysi improv ann learn algorithm improv optim algorithm conjug gradient method learn optim learn system minim error function ann minimum error necessarili gener gener error function measur gener accur imposs practic theori criteria minimum descript length inform criteria minimum messag length criteria defin error function hope minim function maxim gener error function lead gener learn error function bp advanc learn algorithm optim algorithm optim error function natur problem unchang similar situat occur machin learn function defin algorithm minim function error function correspond true gener problem face induct learn method practic good empir function correspond true gener formul learn optim situat justifi evolutionari learn learn method peopl evolutionari algorithm maxim fit function minim error face problem maxim fit function maxim gener evolutionari algorithm optim learn algorithm tradit opportun improv evolutionari learn maximum fit gener evolutionari individu maximum fit popul individu popul inform improv gener learn system benefici make popul singl individu popul inform singl individu combin individu gener form integr system expect produc result paper confirm true conduct set comput studi iii evolutionari design system automat system base evolutionari program design feedforward main structur system shown fig recombin oper simul evolut order avoid permut compet problem reli mutat select scheme evolv architectur connect weight simultan order reduc fit evalu previou studi evolut ann architectur separ evolut connect weight approach introduc fit evalu aim evalu ann architectur ann train weight eventu evalu fit architectur affect random initi weight train train algorithm map genotyp phenotyp common method reduc evalu phenotyp fit fit method gener replac strategi gener similar ga continu ep individu replac offspr gener advantag replac demonstr comput studi mutat type mutat hybrid node connect connect node addit appli time section explain select mutat hybrid train consist modifi bp simul anneal modifi bp adapt learn rate individu popul sa reduc risk stuck bad local optimum distinct featur hybrid train partial train individu converg train fix amount time gener fix amount paramet user motiv partial train comput effici fit evalu train individu converg gener close behavior link parent offspr hybrid train attempt architectur mutat larger behavior architectur mutat partial train phase gener start architectur chang weight adjust yao popul inform evolutionari artifici neural network fig main structur dramat chang behavior partial train phase plai import role maintain closer behavior link parent offspr point train plai role local search ep conduct global search fit evalu base local optimum individu find start state individu poor due architectur mutat node delet total node select uniformli random delet architectur mutat uniformli random connect delet addit nonuniform probabl distribut decid connect delet add base import connect node addit achiev split exist node introduc random node obtain split exist node connect exist node weight node weight vector exist node weight vector mutat paramet fix random node connect node method greatli maintain behavior link parent offspr reduc blind random node improv gener explicit complex term fit function penal larg due difficulti balanc coeffici accuraci complex attempt aic razor balanc term encourag small order mutat order attempt mutat hybrid node connect connect node addit delet attempt addit delet offspr worst individu current addit appli order impli small network prefer job studi benchmark problem shown effect method valid set improv gener evolv train individu train evalu valid set fit valu calcul base train set simul individu gener train modifi bp combin train valid set valid set ieee transact vol june stop train select individu output studi appli problem data set obtain uci machin learn repositori australian credit card data set problem assess applic credit card base number attribut exampl total output class attribut includ numer valu discret valu diabet data set problem exampl class class attribut data set difficult classifi binar form attribut highli indic type diabet correspond medic condit diabet heart data set predict presenc absenc heart result medic test carri patient data set extract larger set data set origin exampl miss class valu discard leav retain case leav final total presenc absenc heart reduct number class origin data set heart valid set experi fit evalu step train individu minimum error rate chosen final result individu minimum error rate combin train set final result tie individu minimum error combin train set final result final individu test unseen test set data set randomli partit train test set size train test set partit previou suggest benchmark input attribut rescal linear function output attribut problem encod output class method output highest activ design class paramet experi set popul size maximum number gener initi number hidden node mean number hidden node initi individu gener random initi connect densiti tabl size problem result averag run max standard maximum mean probabl connect node constraint feedforward violat initi learn rate number mutat hidden node mean node number mutat connect mean number mutat connect paramet select modest search found sensit paramet tabl ii show result problem run error rate tabl refer percentag wrong classif produc comparison error rate obtain method result competit size measur number hidden node small execut phase hardwar implement tabl iii compar result produc bp network term averag test error rate network size iv popul discuss section previou implement ep optim algorithm minim error rate valid set individu chosen final output rest popul discard individu minimum error rate valid set minimum error rate unseen test set rest popul inform improv gener order integr inform individu treat individu modul linearli combin call combin system ensembl reason linear combin simplic find combin show import popul inform advantag combin yao popul inform evolutionari artifici neural network tabl ii accuraci problem result averag run max maximum tabl iii comparison bp network term averag test error rate network size result expect nonlinear combin method major vote simplest linear combin method major vote output number output ensembl output lowest error rate select ensembl output ensembl case popul individu gener particip vote treat equal result major vote problem tabl iv major vote method outperform singl individu problem major vote differ individu perform individu heart problem treat individu popul equal tabl iv accuraci ensembl form major vote result averag run tabl valu compar individu ensembl form major vote valu calcul base run individu equal import perform poorli due mutat previou gener greatest advantag major vote simplic requir virtual extra comput cost tabl show result compar individu ensembl form major vote level ensembl individu australian credit card diabet problem heart problem linear combin differ individu involv extra comput cost fit inform comput weight individu rank gener weight combin ensembl output sort error rate popul output weight ieee transact vol june tabl vi accuraci ensembl form linear combin method result averag run tabl vii valu compar individu ensembl form linear combin valu calcul base run scale factor set australian credit diabet heart number select modest preliminari experi ensembl output result linear combin method tabl vi result produc ensembl good produc individu tabl vii show result compar individu ensembl form linear combin ensembl individu australian credit card diabet problem level outperform individu heart problem statist linear combin rl algorithm algorithm learn linear combin weight linear rl algorithm pp find weight minim number train exampl desir output summat output ensembl righthand side omit discuss sake minim error respect weight yield equat express matrix form uniqu solut exist correl matrix weight vector found invert multipli method train requir multipl matric rl algorithm pp method determin weight implement rl run perform initi yao popul inform evolutionari artifici neural network fig major step evolut ensembl structur tabl viii accuraci ensembl form rl algorithm result averag run weight initi weight gener random unit matrix result run chosen output rl algorithm result ensembl form rl algorithm tabl viii clear ensembl perform individu problem result combin method produc ensembl rl algorithm recommend algorithm perform linear combin algorithm tabl ix show result compar individu ensembl form rl algorithm ensembl individu level australian credit card diabet level heart problem tabl ix valu compar individu ensembl form rl algorithm valu calcul base run subset popul ensembl previou combin individu gener ensembl interest investig reduc size ensembl test error rate investig provid hint individu gener inform shed light import popul evolutionari learn space subset larg popul size impract exhaust search find optim subset genet algorithm search subset weight subset determin rl algorithm section ga popul size maximum number gener crossov rate mutat rate length paramet chosen implement elit ga major step ga summar fig result ensembl form subset gener tabl ga search subset problem tabl xi size subset evolv clear size reduc averag tabl xii show valu compar accuraci individu ensembl ensembl individu australian credit card diabet problem level individu heart problem level result small number gener experi larg number produc search ieee transact vol june tabl accuraci ensembl form optim subset gener result averag run tabl xi ensembl size found ga result averag run tabl xii valu compar individu ensembl form subset gener valu calcul base run time ga run gener heart problem averag test error rate run improv reveal ensembl individu level heart problem paper show learn optim practic learn problem formul optim evolutionari learn exploit differ make inform popul paper show popul inform individu inform effect improv gener learn system paper introduc evolutionari ann design individu gener final output describ method combin individu gener ensembl result benchmark problem show ensembl gener abil individu spite good result obtain individu method investig paper linear combin involv extra comput cost combin method make fit inform readili comput weight achiev good result method fit evolutionari learn due simplic effect combin method base rl algorithm requir comput produc good result comput time primari linear nonlinear method paper natur evolv modular modul individu popul special made evolut modular evolut modular encourag evolutionari expect improv result encourag modular speciat techniqu fit share automat form speci popul speci specialist deal part complex problem treat modul final system modul evolv specif integr system learn evolv modular system test idea successfulli system step implement acknowledg author grate anonym construct greatli improv paper refer yao structur evolutionari artifici neural proc ieee int conf evolutionari comput pp linear combin neural phd school ind purdu west dec regress averag method varianc reduct extens gener convex measur phd dept brown artifici neural int neural vol pp artifici neural encyclopedia comput scienc kent ed marcel vol pp yao popul inform evolutionari artifici neural network genet algorithm neural survei state proc int workshop combin genet algorithm neural network whitlei ed lo ieee comput soc pp yao evolutionari system evolv artifici neural ieee tran neural vol pp design artifici neural network appl math publish learn algorithm learn architectur weight neural chin adv vol pp adapt filter equal intern error parallel distribut explor microstructur vol rumelhart ed mit pp hush neural ieee signal process vol pp jan learn neural network conjug gradient int neural vol pp scale conjug gradient algorithm fast neural vol pp mathemat theori complex vol pp shortest data vol pp statist model ieee tran appl vol pp wallac tech rep dept comput monash aug artifici intellig simul evolut evolutionari philosophi machin intellig ieee genet algorithm connectionist tech rep comput sci eng dept univ san feb algorithm permut comparison recombin oper neural net structur proc int workshop combin genet algorithm neural network whitlei lo ieee comput soc pp evolutionari algorithm construct recurr neural ieee tran neural vol pp empir studi genet oper genet vol pp yao preliminari studi design artifici neural network proc ieee singapor int conf intellig control june pp studi reproduct gener steadi state genet foundat genet ed san morgan pp fogel evolutionari analysi cybern vol pp model select neural vol pp develop multilevel neural neural vol pp inform criterion optim neural network ieee tran neural vol pp zhang optim neural network genet algorithm complex vol pp set neural network benchmark problem benchmark tech rep für universität sept machin neural statist classif elli algorithm gener radial basi function net classif neural vol pp baldi linear neural ieee tran neural vol pp genet algorithm machin learn dilemma fit share scale proc ieee int conf evolutionari comput pp nich method fit share implicit share parallel problem solv natur vol lectur note comput scienc pp modular proc ieee int conf evolutionari comput pp xin yao receiv scienc technolog china msc north china institut comput technolog phd senior lectur school comput south australian defenc forc academi australia held postdoctor fellowship australian nation scientif industri research organ join publish number paper field evolutionari comput neural network dr yao program ieee ieee ak program chair eighth australian joint confer ai confer simul evolut learn program member intern confer associ editor ieee transact evolutionari member ieee technic evolutionari comput yong liu born receiv msc scienc comput mathemat pursu phd school comput south australian defenc forc australia lectur visit fellow south wale publish number paper intern journal book genet algorithm scienc research interest includ neural evolutionari parallel optim 
ML/2495 6 techniqu telecom pari depart informatiqu rue pari research colloquium decemb abstract present review experi techniqu appli task convert graphem string phonem sequenc report experi carri english word french proper experi support view full advantag huge pronunci dictionari develop project tradit understand classif problem question introduct past larg number techniqu appli task automat written word sequenc phonem area speech tradit approach remain term transcript accuraci prove extrem tediou time consum rule develop mainten concern attempt formal handl transcript proper ai commun long shown deep interest understand model cognit process learn ideal task test trivial larg exampl easili explain commun equal contribut develop techniqu present bird ey view approach explor cope difficult problem clarifi terminolog point learn train data refer align pair orthograph graphem orthograph string map phonem deriv fact techniqu review fall categori phonem produc provid train stage transcript place graphem refer graphem examin sourc phonem target phonem final note approach present articl author point put forward cope close relat stress exist techniqu rule induct formal natur framework express phonolog regular orthograph string step domain techniqu direct induct exampl align pair extract rule easi string map requir insert null symbol string align align pair orthograph form number graphem string extract rule techniqu easi difficult minim set unifi rule share context experi procedur recent develop techniqu handl successfulli rule extract limit rule induct scratch concern symbol learn group attempt made attack transcript task symbol learn tool problem view multiclass cluster graphem classifi class class correspond basi context occur learn machin learn literatur larg varieti method suit larg number effect evalu basi perform transcript task popular techniqu techniqu set classifi object seri attribut iter build classif tree tree cluster object class unknown start empti find attribut predict respect classif valu attribut divid origin set subset process repeat iter termin leaf tree correspond object belong attribut success attempt report simplifi version satisfi result multilingu experi carri simplifi version differ fix priori list cluster hierarchi appli object term surround orthograph strategi correspond fix standard mask extract rewrit import featur sourc graphem graphem tree techniqu build order list contextdepend techniqu previou section percent phonem accuraci low nineti report english data accuraci word approach techniqu stress algorithm test corpu fall categori import point rememb famili exampl strictli statist cluster discrimin function learn close approach differ train system report primarili focu set match graphem import point retriev learn set similar order ground cluster case close relat report result major drawback techniqu match process rapidli extrem time consum size match function incorpor linguist knowledg structur match smart retriev time factor limit applic method small learn sampl examin section integr linguist knowledg fall scope strictli techniqu neural network architectur offer statist paradigm cluster problem express solv nettalk famou attempt neural network nettalk nettalk train standard demonstr skill classifi high accuraci letter phonem class basi graphem context window rosenberg architectur extens result refin test recent report extens nettalk capabl learn data result report tend demonstr architectur limit term perform achiev huge dataset pell report experi word learn fed perform phonem exce fair test addit result addit result network found abil architectur cope problem dutch comparison algorithm perceptron perform basi result transcript task noteworthi trick hint gain addit percent accuraci recurr output phonem ad context graphem examin advantag improv knowledg previou phonem inform make add phonem output string condit product phonem previou product improv prove demonstr section result fruit improv cluster map priori class predict output code underli idea output code learn construct class learn result improv note network close relat stress assign altern approach architectur limit attempt made altern network design luca damper describ syntact neural network architectur adapt cope transcript task interest featur train data underli captur framework graphem phonem depend result stai model review complet mention develop architectur map develop report volum discuss experi tend acknowledg idea architectur provid flexibl power framework statist main regular set draw analog perform cluster task unabl captur finer disturb huge train set research field prove direct remain open term algorithm prematur dismiss model basi todai perform automat transcript proper connectionist architectur todai paradigm develop reason found demonstr model cope irregular proper tend irregular reason view transcript cluster problem lead transcrib word letter letter predict phonem improv correct word readili comput averag number graphem result match system symbol cluster method similar argument church liberman advoc advantag larg pronunci dictionari analog markovian model markovian model refer wide class stochast infer method share common popular model hidden markovian model scheme gener idea approach model string predict phonem function stochast sourc emit observ sequenc orthograph sound theoret model paramet transcript problem resolv comput sequenc state produc observ graphem simpl state phonem model map retriev phonem state sequenc correct estim emiss train algorithm effect note train data statist statist estim model probabl popular speech model test transcript task interest attempt report effect approach demonstr simpl bigram model relat work report model predict stress assign hmm tend overfit train lead poor gener lot care estim probabl unseen sequenc hmm suit cope true dynam speech kind task model techniqu reli strictli techniqu depart radic term object term util make train exampl background object method primarili develop goal model observ human perform pronunci task cognit theori advanc account human abil read unknown word present work defend idea unknown word pronounc basi orthograph analog similar word mental lexicon altern view exist parallel cognit primarili model test main goal research area learn main differ model cluster system extract regular train form statist indic set weight neural network sake cognit learn stage model describ trivial record main part work perform analog procedur result system investig applic hypothesi program learn phase simpli consist rememb train data unknown word present search lexicon perform retriev word share letter common word examin retriev exampl organ lattic search pronunci produc path rank path length frequenc data result report good word difficult compar test set list famili profit refer work sullivan damper report result altern model analog occur orthograph phonem string discuss model dismiss basi poor comput complex phonem transcript report perform suggest tractabl provid framework system support recent advanc refer realiz analog pattern encod network drawn parallel distribut strategi step investig develop simpl inspir work byrd result demonstr applic method larg set exampl integr knowledg system mention success applic combin approach techniqu softwar develop gold cope pronunci proper architectur system similar program morpholog decomposit languag identif graphem phonem transcript stress assign modul specif set encod knowledg domain exampl critic support rule exist evid system build appli rule basi analog word exampl advantag gener idea twofold reli weak theori transcript rule specif written proper gener transcript advantag case base rule failur set rule perfectli accur exhaust safe produc decomposit origin process origin lead transcript set final filter procedur rank transcript global analog measur plausibl solut respect exampl result compar compet altern correct human commerci speech synthesi system score low nineti sophist test procedur gold layer bigram approach develop zue author ground approach transcript classic grammar word word made root turn lowest level hierarchi graphem depend kind symbol gener grammar explicitli map data primarili statist enrich assign probabl product rule probabilist bigram sequenc bigram symbol level hierarchi probabl estim set data predict word perform evalu rank larg amount linguist knowledg fed result good author report word correct test set strictli system shortli present exist reimplement adapt belong famili learn inspir attempt investig model section present preliminari result perform compar ibl softwar develop framework ov andersen paul main differ dec vari context size dec ibl algorithm dec expand learn algorithm main differ algorithm fix priori context algorithm length context achiev correct predict dec simpli build smallest rule tree capabl complet error free exampl learn phase desir properti dec price learn step achiev singl pass learn read maximum context size tree give descript clarifi concept context scheme context disambigu dec refer context scheme told charact expand context rule compar implement program small addit featur statist pronunci produc studi section qualiti score procedur capabl handl produc multipl request featur test articl incorpor context scheme graphem result previou predict context scheme flag add rule iter ang iter ne ang ng iter ang tabl learn pronunci graphem french make minor cope fact orthograph entri learn set graphem null present unknown word test procedur pedagog sketch tabl illustr grow rule tree learn stage learn set word context mean context ad graphem graphem graphem word lead creation rule account pronunci match predict phonem conflict refer context expend flag conflict process repeat train conflict rule remain end iter addit treatment establish output conflict rule statist basi end side rule frequent output case unknown word system maxim context found tree match pick past act eur ak da figur sketch behavior tion graphem lead applic rule correct product phonem simpl origin previou relat work byrd algorithm implement main idea larg portion learn word make pronunci unknown word pronunci simpli past begin end exist major intuit argument support approach fact reli larg segment exist consist problem occur letter letter approach disappear learn set program request transcrib sketch figur add word train pronunci system erron answer past import strategi larg overlap word stuck interest problem add pronounc act tion good candid elabor overlap graphem past filter inconsist similar solut accept overlap size simpli solut smaller pedagog ac tion lead correct act ion lead decid basi frequent elabor score procedur assess priori accuraci intuit account import point produc good substr larg part score number overlap size word pronounc pronounc word produc score maximum overlap part mean control consist interact boundari equal result case bad score import factor rel solut elabor term confid solut possibl qualiti criterion comput number past support total number consist past final score simpli sum give credit overlap size factor vote result final note undesir remain silent face word find memori string cover small word easili smaller learn larger number word pronounc huge embarrass situat seldom situat behavior find word correctli transcrib result test procedur nest present result test randomli extract ten pair data pair consist train set test set algorithm train test remain data note due multipl graphem form train test dataset word remov test set dec rememb exampl fail entri learn word test set compar refer pronunci term global word correct phonem correct comput formula pc total number phonem test total number delet substitut transform phonem string produc system refer string develop divid randomli part word set test procedur remain word set divid block increment ad grow learn set evalu run test paramet fix window context size letter side graphem classifi context scheme set nettalk french configur discov dec higher result present term word obtain process word backward word phonem dec word dec phonem phonem figur experi nettalk corpu simpli mean nettalk dec context perform cyclic ad graphem graphem phonem backward nettalk data figur show clear similar dec system achiev averag phonem correct close word correct interest point behaviour term phonem transcript evalu good reason paradox state number word fail phonem count averag seri test comput accuraci number success divid number attempt total number averag correct word interest compar accur algorithm matric reproduc tabl dec behav program correctli pronounc word fail reflect fact program context scheme advantag past phonem reli wrong wrong wrong tabl matric nettalk test correspond run word phonem dec word dec phonem phonem figur experi french graphem dec fail word dec succe word dec predict advantag algorithm permit produc word test word french test run small portion french result reproduc figur stori ibl algorithm give undoubtedli achiev averag term phonem word correct differ figur previou reflect fact french regular languag english ibl similar fact confirm studi wrong wrong wrong tabl matric french correspond run dec word dec phonem phonem figur assess learn rate larg tabl dec succe fail result diagon matrix confirm differ deepli ibl predict correctli dec fail pronounc disappoint perform explain natur ist work languag graphem form made concaten suffix prefix proper corpu ideal situat nettalk corpu hand larg evalu learn rate algorithm figur report experi subject main point larg proport learn dataset result major improv exampl algorithm due technic test partial result confirm close point algorithm give result wrong wrong tabl matric french correspond run achiev accuraci compar word correct achiev exampl fact final perform slightli interest point number exampl dec word correctli produc word demonstr abil effect addit inform addit lesson experi interest notic end pronounc word test eventu remain matrix dec tabl bring encourag result test set correctli predict method partial hold benefit term perform import understand improv improv refin statist estim assess priori reliabl transcript figur figur show evolut score distribut period train train dataset number word high score left translat curv label curv word feel curv mislead dec percent correct continu learn set word multipl drop test result rel improv dec result figur assess learn rate larg test set receiv score result interest regard famili repres percent word accuraci function score largest train set curv word correct close word score higher repres test predict priori correct accuraci kind curv drawn number past larg overlap averag priori score improv word score probabl error word score incorrect perspect present articl altern algorithm capabl cope problem rule larg exampl belong famili instanc learn closer approach prove achiev data word accuraci close estim test word complet random set proper result demonstr techniqu perform rule base system larg learn set improv method wai investig refin statist score procedur account frequenc remain room improv improv control region past pair list pair phonem string improv reliabl pin stick broken part account phenomena occur segment boundari unabl cope correctli predict addit stress syllabl boundari memori admiss vowel word skeleton provid simpl effect filter overlap past produc word primari stress idea appli symbol techniqu speech synthesi system experi nettalk shown choic good context scheme improv significantli result idea mutual serv adjust context integr version order settl paramet set direct provid fruit corpu manual dynam program procedur align realiz map graphem phonem find wai perform map add consist transcript dec concern refer pell connectionist architectur system european confer speech commun technolog volum page sep ov andersen paul approach transcript proper draft research paper project ov andersen paul user manual softwar project john neural network model read neural comput psycholog john connectionist model spell sixteenth annual confer cognit scienc page aug roi byrd martin dictionari find word pronunci unknown word annual meet associ comput page jul eugen charniak statist languag learn mit kenneth church stress assign letter sound rule speech synthesi annual meet associ comput linguist page jul van coil induct learn pronunci rule system intern confer speech signal process volum page cecil kenneth mark liberman morpholog power altern rule confer speech walter steven acquisit stress approach comput sep walter steven van den learnabl acquisit stress technic report tilburg michael howard pronounc program pronunci analog comput speech thoma dietterich bakiri output code gener method improv multiclass induct learn program meet american associ artifici intellig page thoma hermann bakiri compar studi english map morgan intern machin learn compil larg pronunci pf pattern network complex organ activ orthograph knowledg read aloud journal psycholog human percept principl pronounc print psycholog interact process page erlbaum andrew gold pronounc combin reason phd stanford oct neural network research nov helen stephani victor zue gener base word morpholog european confer speech commun technolog volum page sep deni shipman rule discoveri procedur journal acoust societi kohonen associ memori kohonen dynam expand context applic correct symbol string recognit continu speech intern confer pattern volum page oct wendi lehnert problem solv larg knowledg base learn case meet american associ artifici intellig page luca robert damper syntact neural translat gerard eric talk page north jm rl mercer inform theoret approach automat determin phonem intern confer speech signal process volum page san helen stephani victor zue phonolog gener intern confer speech signal process volum page apr raymond geoffrei alan comparison symbol algorithm intern join confer artifici intellig volum page aug reimplement nettalk comput speech rc induct learn pronunci rule hypothesi test correct intern join confer artifici intellig page aug sh ra model english pronunci european confer speech commun technolog volum page sep pl transcript rule stress assign technic report ross quinlan learn effici classif procedur applic chess end game jaim tom machin learn chapter page tioga publish palo pa gk phonem graphem hmm european confer speech commun technolog volum page sep interact pronunci rule lexic read aloud memori craig stanfil reason appli english pronunci meet american associ artifici intellig page sullivan robert damper approach pronunci system intern confer speech signal process volum page apr sullivan robert damper speech synthesi analog recent advanc result intern confer speech signal process volum page sullivan robert damper investig german english intern confer spoken langag process volum page oct sejnowski tj rosenberg cr network learn pronounc english text complex combin neural network transcrib speech phonem volum page nov effici learn english rule automat intern confer speech signal process volum page apr van den walter daeleman gener perform backpropag learn task intern confer artifici van den walter daeleman method european chapter associ comput page 
AI/529 2 behavior histori learn nonstationari condit francoi michaud maja matar depart electr comput universit de comput scienc southern lo ca abstract robot oper nonstationari condit learn modifi control polici adapt chang dynam environ approach manag interact robot method model interact adapt select behavior histori behavior learn valid pioneer robot context forag task result show effect approach advantag regular experienc lead fast adapt special learn robot introduct robot oper situat requir abil adapt chang dynam nonstationari control polici chang time mobil robot design paradigm shown good perform adapt oper open environ approach robust simplic construct typic constitu behavior design execut priorit fix flexibl arbitr mechan fix parsimoni behavior robot easili adapt result diminish flexibl improv dynam select behavior extern intern state control polici modul control select mechan work abstract behavior preprint submit elsevi preprint februari import modul nonstationari condit requir learn model environ robot chang dynam environ modifi polici behavior select level deriv model challeng problem care made select behavior nonstationari independ robot oper environ limit action process perform situat agent base abil cope exploit dynam interact environ model interact consist control mechan case framework preserv kei strength emerg model interact dynam environ case dynam environ topolog metric model directli appli dynam difficult design fulli predict captur robot idea explor paper interact robot environ effici captur observ behavior time robot initi set subset safe accomplish assign task goal algorithm learn select altern behavior exploit order adapt robot behavior experienc dynam robot autonom evalu model type interact environ base robot histori behavior regular sequenc behavior reflect interact robot world robot deriv model reinforc learn fashion perform criterion deriv behavior algorithm continu adapt learn static polici approach valid pioneer robot context forag environ paper organ section describ setup mobil robot show behavior organ deriv interact model chang robot behav section detail explan learn algorithm section present result section explain properti learn algorithm relat nonstationari condit section present relat field reinforc section conclud paper setup task descript experi perform real world interfac pioneer mobil robot figur equip sonar fast track vision system robot program mar agenc reactiv languag program multipl concurr process behavior experi conduct rectangular pen pink block home mark green cylind color object perceiv robot vision suffer variat light intens locat pen angl approach obstacl wall detect sonar dark stripe wall pen strip cardboard improv sonar reflect robot equip contact perceiv obstacl pioneer sonar fig pioneer robot home region robot forag search block bring home region pioneer set behavior initi control govern subset behavior behavior control veloc rotat robot base sonar read visual input analyz interact deriv observ behavior behavior classifi respect robot behavior accomplish behavior search call home home behavior call task make robot move behavior refer condit activ base presenc absenc block front robot intern variabl indic robot block underneath field view push proxim robot home avoid behavior requir safe navig environ activ robot home carri disabl order robot approach home region type handl harm situat accomplish refer design determin condit occur ti dynam interact robot environ robot behavior introduc varieti action repertoir chang accomplish task call contrast type priori condit activ object robot learn activ behavior past experi forag pioneer decid activ period time experienc accomplish task figur show architectur specif compon experi behavior behavior organ subsumpt architectur differ activ issu chang dynam behavior activ particip control robot actual behavior depend rule command issu behavior subsum behavior design mutual function emerg activ multipl behavior time home learn adapt behavior select figur show behavior observ learn interact model behavior assign symbol wire send messag behavior execut gener symbol sequenc repres behavior process sequenc construct tree robot interact environ activ separ learn tree determin base activ learn algorithm home avoid rest command interact model level behavior level behavior sequenc vv activ tree histori behavior fig organ behavior level interact model level color background repres behavior activ search task chosen sensori input behavior shown note behavior classif independ subsum organ experi tree mechan algorithm tree structur store histori behavior upper part figur show typic tree node store type behavior control robot accomplish number time transit node successor made observ behavior node type correspond predomin behavior type node repres search task type node creat home task behavior exploit consecut synchron root node tree observ start task letter behavior node type behavior sequenc shown figur path creat tree sequenc home task tree home behavior search task initi observ behavior sequenc current node tree node depict bold tree task empti increment construct robot task leaf node letter store total perform tree path path complet sequenc behavior averag store current perform trial number determin empir comput store histori behavior result perform valu provid abil learn trigger dynam observ robot environ metric perform depend domain varieti factor evalu perform metric learn algorithm chose studi experi base time behavior object idea learn observ behavior robot limit capabl oper dynam environ optim difficult defin nonstationari detail justif choic section evalu express relat amount time requir accomplish task experienc max tb mb tb max tt tt tb refer mb current cycl evalu activ tt repres averag total number accomplish task updat complet task number determin empir analyz typic establish adapt environ toler poor perform result explor option tt captur dynam environ affect time complet task correct factor boost robot term relat reflect amount control robot task compar effect event handl term penal robot perform time accomplish task past note shown figur sequenc behavior situat robot regard past amount time singl behavior tempor separ node consid effect decid behavior activ memor end node trial activ criteria behavior select tree evalu algorithm basic make activ behavior set option refer observ activ sequenc node tree path interact experienc robot situat robot rel interact select criteria compar perform current posit tree expect perform expect perform node sum store perform multipli frequenc rel current posit tree expect perform start posit tree relat refer access perform memor number time node visit expect perform option current posit tree formula determin expect perform simpl bold node tree figur current node expect perform start node expect perform total expect perform node note longer path necessarili lower evalu function influenc number node path time spent behavior expect perform current posit tree choic made algorithm base ffl option maximum expect perform criterion refer tm maximum ffl situat untri option select untri option select criterion refer untri option untri option perform trial option obtain global option tabl arrai past perform index option updat end trial averag result perform option trial evalu behavior activ criteria place mainten node node creat observ occur accomplish robot result node select directli set lower tm greater equal condit posit overrid select made tm note path expect perform default option behav robot explor accomplish path delet mechan algorithm noisi nonstationari delet path interact model tree repres date expect perform weight averag base frequenc occurr option greater influenc variat option make harder modifi chang strategi option forget path lead explor choic discov world node delet serv regul memori enabl robot respond recent opt delet oldest path tree path store criteria easili substitut applic experi goal algorithm learn dynam inappropri start directli domain experi nonstationari condit make analysi valid properti algorithm extrem difficult gradual address static environ condit complex verifi algorithm learn stationari condit experi multipl robot perform behavior repertoir set optim retrain vision system paramet color detect object creat situat design adjust behavior priori knowledg environ learn occur trial search block bring home note learn robot start task search task start drop block direct home depend proxim robot wall posit drop block rel home task start block direct robot face block comput limit amount memori entir robot program kbyte kbyte learn leav kbyte store tree node approxim fig environ configur pen divid region simpler static unexpect behavior select learn robot configur singl block center wall corner good strategi easili find block robot learn rest front wall order rotat avoid behavior exploit block directli locat learn averag total number accomplish search task compar cycl learn algorithm find frequent thought behavior robot move center pen block behavior made robot rotat random angl random reliabl robot block frequent made robot loop block import note discoveri made robot point design perform experi start block center divid pen region figur figur show plot junction tree number node tree learn trial experi robot learn behavior follow wall block chang strategi chang observ option rest trial observ rest option junction tree chang environ configur occur chang strategi stabil start trial corrobor figur demonstr algorithm quickli adapt environ approach show transfer learn configur initi strategi serv bia elabor special prevent chang dynam end averag total number accomplish search task similar result obtain robot learn pen divid region configur show algorithm improv rapidli adapt behavior select strategi situat experienc environ algorithm converg stabl polici reveal optim regard robot perspect includ limit perceptu field detail descript result obtain static condit present block brought home put center shown figur robot ad gener environ program move avoid obstacl stai forbidden contact mount front gripper forbidden region area pen mark dark tile area surround home corner pen control detect obstacl forbidden robot move backward turn move forward junction trial number number node fig graph number node search task tree experi start block chang divid pen region move forward freeli constraint motion result great deal dynam dark area inaccess sensor enabl enter area nonetheless fig environ configur pioneer program avoid obstacl area mark dark tile tape fifteen experi trial hour learn pioneer robot compar learn strategi addit sourc affect behavior face sonar pioneer robot search block perceiv pursu block push blue pioneer robot red pioneer robot pink block nonstationari condit presenc movement vari greatli experi stai close home move region block learn primarili depend sequenc past experi optim determin task larg amount data statist counter data studi experi video tape order analyz strategi regular found interact algorithm adapt detect interact describ result obtain search task illustr properti learn algorithm learn strategi larg involv rest option sequenc option learn observ observ rest rest rest strategi chang time past perform strategi base learn stabl strategi experi fig search task tree experi figur repres part tree learn robot trial complet tree made approxim node observ option node mainten node observ trial posit equal maximum posit tree note path depend select option path time trial figur show trace choic made trial trial test data choic delimit consist trial option select criterion posit tree tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm tm fig trace choic made search task tree experi choic made initi favor option strategi remain stabl trial end observ rest strategi prefer trial option explor choic made tm criteria consecut choic valu made posit tree notic trial choic tm criterion overridden made criterion fell shorter path sign robot find block experienc longer sequenc choic gener longer path figur node algorithm robot discrimin situat observ rest rest front static obstacl rest front move obstacl distinct made behavior special process distinguish move static lead distinct strategi behavior select figur show number node tree stop start demonstr store path tree nonstationari condit pioneer robot put pen special differ wai accomplish base number node tree search task trial number fig graph number node search task tree experi specif experi test strategi learn pioneer search task observ rest observ rest learn tree node node frequent path happen pioneer interact greatli easier opportun find block pioneer pen influenc pioneer abil find stabl dynam interact environ effect dynam special occur requir complex commun arbitr scheme manag group discuss experi valid properti algorithm involv import properti nondeterminist nonstationari environ dynam chang agent experi dynam condit determin present state environ complet determin current state action select agent access percept learn robot limit sonar color inform posit posit robot configur environ effect action state occas result state make ariti difficult explain definit relat learn algorithm al environ stationari probabl make state transit receiv specif reinforc signal chang time sutton barto explain nonstationari mean true valu action actual expect reward action chang time chang reinforc signal experi creat nonstationari condit occur environ modifi polici chang valid experi static condit start block center continu pen divid polici learn configur valid occur experi distribut robot pen stationari condit move object pen remain uniformli distribut environ effect learn robot case start evenli distribut chang posit move center pen block move home region lot difficulti learn robot deposit move remot area learn robot interf behavior select polici continu adapt base approach share similar differ reinforc learn paradigm learn manner similar mont carlo learn algorithm averag observ return eventu converg expect learn deriv store past experi current situat action form behavior reward base intern measur deduc behavior time reinforc learn method sensat state command action sutton barto reason restrict state typic applic expect state inform agent approach dynam program approach aim find tabl maxim util valu behavior serv abstract state behavior action behavior process optim fix util function state base behavior combin environ sensori inform control polici behavior depend activ behavior activ made interact model similar combin control signal newli activ sensori input influenc control signal issu newli activ behavior possibl behavior subsum follow principl emerg approach activ behavior control robot accord experienc construct deriv behavior kei differ commonli reinforc learn algorithm discret reinforc complet determin perceiv condit evalu function experi differ approach attempt optim extern criteria time requir accomplish total distanc number mahadevan kaelbl crucial approxim function reinforc learn continu larg finit state space problem perform function relat enabl learn algorithm make learn independ constraint limit robot specif design behavior exist effici strategi approach aim effici behavior accord robot abil interact environ fix extern evalu learn perform abil approach make consist base past good balanc explor exploit option result demonstr algorithm abil approach gener unanticip behavior select burden behavior design behavior acquir knowledg specif complex dynam system robot learn strategi difficult design anticip algorithm attempt captur dynam experienc base tree evalu function compon learn algorithm establish explor adapt exploit stabl behavior select strategi evalu function current situat past experi tree captur sequenc behavior compact fashion make base past experi behavior activ criteria exploit compon determin choic tm criterion favor exploit criteria favor explor find option plai intermedi role explor option node algorithm choic exploit option perform serv memori perform option make learn algorithm adapt easili environ reflect constant fashion perform trial frequent difficult chang evalu made larger number trial subject faster explor exploit outcom recent trial criterion prefer random behavior select favor knowledg deriv past experi blindli explor option path delet influenc choic made affect tree delet factor defin memor abil view principl bia set design affect type strategi learn result continu adapt behavior requir extens train period model nonstationari condit influenc algorithm establish critic point behavior factor activ behavior critic point modifi continu alter behavior variabl robot abil due batteri nonstationari critic point provid robot abil make expect base previou evalu current perform relat work approach fit categori hierarch method deal larg state space learn involv structur gate collect gate function base state behavior action switch execut work gate function fix learn occur behavior suppli individu reinforc function work involv learn gate main differ approach histori behavior percept state select behavior compar learn approach approach tackl problem hidden import mobil robot equip limit noisi sensor pomdp limit markov properti state maintain belief state form probabl distribut theori reinforc learn model exact solut pomdp gener consid present difficult optim task requir exponenti amount space time comput polici requir exponenti size solut pomdp ten state minut converg obtain shown exact solut problem intract cassandra al pomdp model make mobil robot navig real offic environ learn pomdp model valid static environ condit learn pomdp model koenig simmon make approach subset probabl constant made ident initi backward probabl variabl approxim slide time window approach probabl observ insid base algorithm learn histori past experi inspir mccallum approach partit state space raw sensori memori featur encod augment agent perceptu input featur label branch perceptu dimens data histori index step leav tree store util estim action mechan successfulli appli learn drive virtual highwai environ approach learn window serv simultan model environ polici algorithm learn strategi behavior behavior abstract mccallum approach learn complet polici discret state space good concern arrang behavior control robot independ discret continu sensori state work system inform discrimin state focu learn thing priori improv learn time addit appli perform measur independ task concern approach differ mccallum select criteria algorithm path maximum expect reward random algorithm compar current expect perform tree explor affect path delet statist evalu leaf node learn observ behavior activ explor reason work ram santamaria comput complex approach greater develop algorithm context gener control architectur intellig agent base dynam select behavior learn nonstationari challeng problem interact environ affect dynam robot limit approach advoc intern evalu interact advantag valid approach dynam observ behavior select mechan consist framework rel exploit control resourc defin robot perspect result show effect approach advantag regular experienc world fast adapt agent gener condit result show approach lead learn unanticip behavior strategi perspect robot robot autonomi approach predict capabl typic lack make move abstract level gener complex behavior acknowledg support michaud provid nserc scienc engin research council research fund offic naval research grant nation scienc foundat infrastructur grant career grant matar author goldberg rest interact lab insight comment refer dynam structur everydai phd massachusett institut coordin multipl behavior acquir reinforc proc intl conf intellig robot system robust layer control system mobil ieee journal robot intellig artifici multipl agenc reactiv technic robot pack act discret bayesian model proc intl conf intellig robot system dorigo robot develop autonom agent artifici floreano evolut home navig real mobil ieee tran goldberg tool design evalu proc nation conf artifici intellig pack reinforc journal artifici intellig koenig learn probabilist model robot proc ieee intl conf robot autom dynam action proc intl joint conf artifici intellig mae learn coordin proc nation conf artifici intellig vol mahadevan automat program robot reinforc artifici mahadevan pack nsf workshop reinforc summari ai magazin exampl group journal theoret artifici reinforc learn autonom learn select attent memori sequenti anim proc intl conf simul adapt mit reinforc learn select percept hidden phd depart comput ct le control architectur combin motiv situat autonom anim proc intl conf simul adapt mit michaud learn histori mobil robot machin learn autonom joint special issu learn autonom parr approxim optim polici partial observ stochast proc intl joint conf artifici intellig ram multistrategi learn reactiv control system autonom robot machin handbook percept vol chap russel artifici intellig modern learn partial observ markovian proc intl conf machin learn sutton reinforc introduct 
IR/2180 5 document word bipartit spectral graph partit ut cs technic report tr depart comput tx usa march abstract document cluster word cluster import problem vector space document collect repres matrix present idea model document collect bipartit graph document word pose cluster problem graph partit problem give spectral algorithm simultan yield cluster document word algorithm left singular vector appropri scale matrix yield good bipartit shown singular vector give real relax optim solut graph bipartit problem present result verifi result algorithm work practic robust presenc introduct cluster group similar collect unlabel document cluster organ collect facilit futur navig search document cluster wide studi problem algorithm task start point appli cluster algorithm document collect creat vector space altern model basic idea extract uniqu word set document treat word featur repres document vector weight word frequenc featur space entir document collect treat matrix row correspond word column document entri ij presenc word document entri absenc larg number word exist moder size set test case word document document typic small number word typic matrix entri basic word document cluster algorithm algorithm di er mathemat model ecienc scalabl size collect cluster import variabl cluster larg portion entir world wide web di erent proposit cluster smaller document collect return web search engin exist document cluster method includ agglom partit project base method includ selforgan comput ecienc requir hybrid approach consid recent activ document techniqu consid earlier hierarch agglom cluster recent model similar document graph vertic correspond document weight edg hyperedg give similar vertic major drawback method work requir form graph quadrat number method prohibit larg document collect word cluster basi document cluster automat construct statist thesauru enhanc underli assumpt word typic similar concept word cluster pro automat classi cation detail treatment word cluster techniqu problem simultan document word exist work document word cluster common theme exist algorithm cluster document base word distribut word cluster determin document point dualiti document term cluster pose dual cluster problem term nding minimum cut vertex partit bipartit graph document word find global optim solut graph partit problem show left singular vector suitabl normal matrix give optim solut real relax discret optim problem base present spectral algorithm simultan partit document demonstr algorithm good global solut give outlin paper section introduc bipartit graph model repres document pose problem nding minimum cut bipartit graph constraint size cluster make graph partit problem section introduc spectral graph partit heurist show eigenvector gener eigenvalu problem optim solut real relax graph bipartit object section show bipartit dual cluster obtain left singular vector suitabl normal matrix section algorithm addit singular vector directli yield multipl document word cluster section present detail result demonstr algorithm good solut practic discuss relat work section section present discuss futur work word letter denot column letter denot script letter denot vertex set bipartit graph model introduc relev terminolog graph graph set vertic set edg jg edg weight ij adjac matrix graph de ned ij edg partit vertex set subset cut plai import role paper ij de nition cut easili extend vertex introduc bipartit graph model repres document collect undirect bipartit graph tripl fw wm fd set vertic set edg dg correspond set document correspond set word document edg fw word occur document note edg undirect bipartit edg word document edg signi es associ document word put posit weight captur associ possibl equal term number time word occur document formula inform retriev detail popular scheme ij edg fw ij ij log jdj jd ij number time word occur document jdj total number document jd number document word matrix ij equal ij easili adjac matrix bipartit graph written order vertic rst vertic index word index document show cut di erent vertex de ned emerg natur formul word document cluster simultan cluster basic algorithm observ dualiti word document word cluster induc document cluster document cluster induc word cluster document cluster word cluster determin word belong word cluster wm associ document cluster dm greater associ document cluster graph natur measur associ word document cluster sum document cluster wm ij ij word cluster determin document cluster similarli word cluster induc document cluster dm ij ij note natur document cluster determin word turn determin document cluster word document cluster correspond partit graph cross edg partit minimum weight achiev min partit bipartit graph vertex subset note paragraph hint algorithm formul naiv typic lead unbalanc cluster size minimum cut obtain vertic belong partit cluster mechan constrain size cluster pose object function favor cluster addit graph partit graph classic graph bipartit bisect problem vertex subset min graph partit problem import circuit partit vlsi telephon network load balanc parallel reorder problem heurist method exist local minimum problem ectiv earlier heurist algorithm algorithm emploi heurist search local vicin initi partit tendenc stuck local minima spectral graph bipartit spectral graph partit ectiv heurist introduc earli popular simon spectral partit heurist eigenvector matrix bipartit graph found give good global solut solut obtain kl fm algorithm introduc spectral partit heurist graph vertic edg incid matrix denot matrix row vertex column edg column edg jg ij ij ij edg weight note ambigu de posit posit neg entri arbitrari ambigu import de nition laplacian matrix lg symmetr row column ij ik ij edg jg theorem laplacian matrix lg graph properti adjac matrix diagon matrix ii ik symmetr posit nite matrix eigenvalu real full set real orthogon eigenvector vector le eigenvalu eigenvector graph connect compon eigenvalu equal vector lx ij vector scalar proof part de nition easili multipli lx impli symmetr posit nite matric nonneg real eigenvalu full set orthogon vector lx row correspond edg easi ij le equat part tu rest assum graph consist connect smallest eigenvalu laplacian nonzero part eigenvalu eigenvector give inform partit graph bipartit de ne partit vector captur theorem show quotient proport cut induc partit vector theorem laplacian matrix partit vector quotient lp proof part theorem lp ij edg contribut edg contribut time tu eigenvector optim partit vector theorem cut minim trivial solut cut captur strength associ di erent partit object function addit small cut valu captur cluster present object function vertex posit denot diagon matrix ij subset de ne weight ii subset respect weight equal object function favor balanc di erent partit cut object function smaller balanc partit minim favor partit balanc addit small cut show quotient gener partit vector equal object function lemma graph laplacian vertex weight matric gener partit vector element sati es proof ii similarli show ii tu theorem notat lemma lq proof easi show gener partit vector written partit vector part theorem lq substitut valu lp theorem lemma prove result tu global minimum solut object function restrict attent gener partit vector form lemma problem real relax discret optim problem theorem problem min lq wq subject solv eigenvector smallest eigenvalu gener eigenvalu proof standard result linear note rst eigenvector condit constrain search optim vector rst eigenvector tu theorem impli eigenvector real approxim optim gener partit vector corollari corollari smallest eigenvalu lower bound object function object speci ed choic vertex weight simpl choic vertic lead object consid circuit jv jv interest choic make weight vertex equal sum weight edg incid ik lead criterion imag segment note choic vertex vertex weight matrix equal matrix sum weight edg object function express ik ik note measur strength associ partit minim maxim proport edg weight lie partit spectral bipartit graph partit svd previou eigenvector gener eigenvalu real relax discret optim problem nding minimum normal cut present algorithm document word cluster bipartit graph model bipartit diagon matric ij incid word ij incid document written ay assum document word word document rewrit equat ay algebra ad equat de ne singular decomposit normal matrix ad left singular vector singular comput eigenvector eigenvalu comput left singular vector singular work superior size matrix larger size singular vector give bipartit document left singular vector give bipartit word examin relat clear solut intuit partit document induc partit partit word impli partit document bipartit algorithm singular vector give real approxim discret optim problem minim normal cut kei task extract optim partit vector optim gener partit vector lemma strategi distribut valu denot valu previou eigenvector approxim optim bipartit assign valu criterion object function classic algorithm algorithm word document algorithm bipartit form ad comput singular vector form vector run algorithm data obtain desir bipartit aspect algorithm run simultan reduc word document section algorithm yield good result practic algorithm adapt bipartit algorithm gener problem nding word document cluster solv problem algorithm bipartit manner favor direct approach singular vector observ collect dlog ke singular vector inform data set form data set dlog data point assign minim classic algorithm obtain algorithm algorithm form ad comput dlog ke singular vector form matrix run algorithm data obtain desir result popular cranfield document set medlin consist abstract medic consist abstract inform retriev cranfield consist abstract aeronaut system paper document collect download test algorithm creat data set mixtur collect creat document medlin cranfield collect creat matric data typic remov stop word occur descript medlin word occur document remov medlin word stop includ medlin word occur document remov medlin word stop includ cranfield word occur document remov cranfield word stop includ cranfield word occur document remov cranfield word occur document remov cranfield word occur document remov cranfield word occur document remov yahoo reuter articl word stem heavili prune yahoo reuter articl word stem stop word prune tabl descript data set greater document word occur infrequ prune step drastic reduc number word prune strategi remov total word show algorithm robust presenc larg number word matric obtain includ stop word detail test collect tabl test algorithm creat data set mix cranfield total document show algorithm work small data creat subset document nal data set collect reuter articl download yahoo octob articl sport articl technolog data set preprocess elimin stop word html tag word stem porter data set yahoo word yahoo includ word obtain remov stop word tabl detail document word nonzero yahoo yahoo tabl detail matric bipartit result present bipartit result collect class label matrix captur good document cluster measur puriti entropi class class cluster algorithm produc cluster cluster puriti de ned max jd number document belong class cluster sampl di erent class puriti ratio domin class size cluster cluster size high puriti mean cluster subset domin class entropi qualiti de ned log log entropi measur puriti consid distribut class cluster note normal entropi valu entropi mean cluster entropi bad impli cluster uniform mixtur class tabl result appli algorithm bipartit data set matrix top left tabl show document cluster consist medlin identi ed cranfield puriti cluster entropi close medlin cranfield puriti entropi patient cell blood children hormon cancer renal rat cell growth shock heat superson wing transfer plate jet tabl bipartit result medlin puriti entropi cell patient blood hormon renal rat cancer cell len dna librari retriev scienti research scienc system book comput index journal tabl bipartit result bottom tabl displai word word cluster top word intern edg weight greatest note algorithm bipartit document word cluster document cluster observ top word convei document cluster tabl show good bipartit obtain spectral algorithm data set algorithm bipartit global spectral heurist singular vector make robust presenc word show ran algorithm data set obtain remov stop word matric tabl show algorithm recov origin class presenc stop word note medlin cluster pure case puriti entropi librari retriev scienti scienc book system index research literatur journal layer heat shock mach superson wing plate tabl bipartit result medlin cranfield medlin tabl matric med puriti entropi patient cell blood hormon renal cancer rat cell len librari librari retriev scienti scienc book system system research index boundari layer heat shock mach superson wing pressur tabl result result show algorithm good result multipl word cluster requir tabl entropi document cluster top word word cluster note algorithm dlog ke singular vector yahoo yahoo data set class sport technolog entertain domin class document technolog articl size class vari tabl result obtain dlog ke singular vector dicult recov origin class presenc matrix encourag tabl show cluster identi ed entertain health sport word cluster show underli concept document cluster word stem tabl show similar document cluster obtain fewer word algorithm small collect tabl show mix small subset cranfield algorithm recov class stark contrast spheric algorithm poor result small document relat work mention section word document cluster problem work address dualiti document cluster word cluster problem studi bu entertain health polit sport tech puriti entropi clinton campaign senat court nanc white compani reform stock septemb tv week music set top fridai record lm star award comedi henderson world health polit entertain tech sport index surgeri injuri undergo hospit england accord recommend headach heart republ advanc wildcard match tabl result yahoo bu entertain health polit sport tech puriti entropi compani stock nanci pr busi wire quot percent industri gain lm tv comedi previou entertain adult albert presid washington bill court militari octob violat trade appeal secretari health pm death famili rate lead driver peopl surgeri injuri undergo hospit england recommend discov accord heart cau senat clinton campaign white nanc vote tabl result yahoo med med med tabl matric gener data earlier work limit matric small dimens method spectral bipartit partit cluster word document normal cut criterion imag segment eigenvector singular vector document cluster recent lsa method project document vector higher subspac prohibit algorithm intim relat inerti rst princip compon cluster document normal scheme natur normal cut criterion enabl left singular vector bipartit di erent normal lsa result obtain poorer matric tabl point show optim spectral partit scheme variant techniqu theoret shown give partit close optim special class planar recent work obtain document cluster graph work similar graph vertic correspond document edg correspond document similar obtain similar measur cosin gener jaccard similar method prohibit form similar graph requir work quadrat number document futur work introduc idea model document collect bipartit graph word document pose problem word document vertex partit problem bipartit graph solv graph partit spectral partit heurist shown left singular vector suitabl normal matrix give real relax discret optim problem multipl document word cluster obtain addit singular vector result spectral algorithm global good solut robust presenc word result present paper algorithm futur conduct detail experi larger document collect compar ecienc ectiv document word cluster algorithm improv spectral partit heurist plan add avenu explor method bipartit graph partit minim criterion current import return equal size optim di erent object function accept integ web bipartit graph base link enhanc web search futur plan combin word link uniform manner cluster web document main contribut paper bipartit graph model text collect successfulli appli appli bipartit model text mine text classi queri fascin aspect bipartit graph vertex word follow zipf subgraph fractal futur intend studi properti graph feel algorithm dna gene express partit rectangular matric load refer dougla baker andrew mccallum distribut cluster word text classi cation sigir annual intern acm page august obrien linear algebra intellig inform retriev siam bolei hierarch taxonomi partit technic report depart comput moor document categor queri gener world wide web webac ai cheng wei improv partit algorithm stabl perform ieee transact decemb approach thesauru construct intern confer research develop inform page june tukei approach larg document collect acm concept decomposit larg text data cluster machin januari ibm research report rj juli ho man lower bound partit graph ibm journal research stork pattern classi cation john wilei edit linear time heurist improv network partit technic report gener electr corpor research develop algebra connect graph mathemat garei johnson comput guid theori freeman gene golub charl van loan matrix comput john hopkin studi mathemat scienc john hopkin hagen spectral method ratio cut partit cluster ieee transact hall quadrat placement algorithm manag hartigan direct cluster data matrix journal american statist march hartigan cluster algorithm hearst pedersen reexamin cluster retriev result acm page karypi kumar fast high qualiti multilevel scheme partit irregular graph siam sci samuel timo kohonen map document collect studi document scale index term system develop santa bw kernighan lin ecient heurist procedur partit graph bell system technic jon kleinberg authorit sourc hyperlink environ symposium discret extend version journal acm ibm research report rj kohonen map matrix method applic phd appli mathemat colleg kruskal numer method alex partit matric eigenvector graph siam journal matrix analysi juli rasmussen cluster algorithm william frake ricardo inform data structur page prentic englewood cli baldonado servic organ network inform autonom digit librari acm confer digit page salton bucklei approach automat text retriev inform process salton mcgill introduct modern retriev book silverstein project ecient document cluster acm shi malik normal cut imag segment ieee tran pattern analysi machin august andrew simon scalabl parallel dynam adapt comput supercomput daniel spectral partit planar graph nite element mesh annual symposium foundat comput page octob moonei impact similar measur cluster workshop arti cial intellig web page juli press van rijsbergen inform retriev ectiv ecienc agglom hierarch cluster document retriev phd cornel willet recent trend hierarch document critic review inform process william uni cation spectral inerti bisect technic report center advanc comput california institut zamir etzioni web document feasibl demonstr acm zipf human behavior principl ort addison 
Agents/1926 1 autonom agent fulli embodi make commun behavior autonom cassel mit media ma usa avatar resembl commun interfac part profit recent research autonom embodi system avatar function environ chat resembl human incap repres kind knowledg human bodi commun make extens visual channel interact manag subtl involuntari cue read gaze gestur argu model anim fundament behavior crucial credibl effect virtual interact chat treat avatar commun method autom anim import commun deriv work theori system user commun text avatar automat anim turn feedback facial express evalu show user found avatar autonom behavior natur avatar behavior perceiv express user felt avatar autonom commun behavior provid greater user control embodi commun behavior behavior avatar type embodi agent receiv research attent agent avatar graphic chat avatar repres user distribut virtual autonom knowledg act absenc explicit control part user current graphic chat system user oblig switch control avatar behavior type messag user user creat messag avatar stand repeat select anim sequenc fail reflect natur relationship bodi potenti give mislead conflict visual cue user system offer simpl lip greatli enhanc action gaze gestur incorpor simpli produc random creat develop graphic chat environ irc awar part design import commun modal creator environ realiz avatar anim order bring approach account number kind commun function bodi encount provid menu user select set anim sequenc switch emot largest problem approach user explicitli control chang avatar state realiti visual cue import spontan make imposs user explicitli select menu user busi produc content simultan behavior control burden design stiff earli version avatar consid wai make gener lack emot live emot express interact mechan establish maintain mutual focu attent place tend commun behavior gaze head movement spontan natur fluid execut make easi overlook recal previou encount creat avatar humanoid agent emot displai account major displai occur human human interact approach fulli begin develop theori embodi agent reli find literatur analysi function commun behavior ey head eyebrow human human mediat machin role interfac gener accord theori embodi agent account function user prefer direct manipul agent autonomi avatar kind agent design problem integr research find commun behavior issu interfac design interest avatar repres commun priori intent commun report research call envelop role support experi user system art avatar term avatar refer wai repres network user graphic rang applic broad requir user virtual presenc differ work implement type avatar inhabit technic refer distribut virtual environ activ world browser fire studio commun client oz virtual exampl system idea present applic kind system view mind popular system busi system place avatar portrait top flat imag backdrop render virtual environ basic interact principl give idea current state art shortcom current descript typic popular commerci order client program connect user server maintain state share world client render view world avatar ey camera float avatar avatar articul model user menu bodi construct suppli editor articul includ arm feet motion rare facial express gaze movement virtual provid facial user freeli navig avatar scene cursor kei user type sentenc edit transmit world hit return text window directli render view displai transmit sentenc user sentenc float head user avatar user initi contact step user navig attempt enter person field view user select set anim sequenc avatar wave situat user start transmit sentenc prefer address person contact step user greet sentenc heard avatar virtual room avatar exact locat orient user type messag switch avatar anim set jump wave select anim idl stretch check watch randomli execut entri world notic live fact world crowd peopl gather citi squar crawl avatar move stretch bodi realiz anim displai reflect actual event transcrib scroll text window beneath world view avatar user visual creat format control posit orient relat affect user abil commun long desir audienc virtual room reason redund bodi system convei signal autom motion sequenc link state content initi make irrelev manual execut motion sequenc explicit emot chosen user button control tend user engag type keyboard typic session paul walk susan stand stare space susan watch repli great im paul return stare limb life dont im back susan watch paul continu walk random person start susan watch karen paul rotat full circl susan repli karen gui karen virtual mile wave susan paul great peopl talk paul stiff karen wave susan watch approach attempt amelior unnatur experi explor fulli integr behavior bodi phenomena make avatar behavior autonom emploi tracker map kei part user bodi face graphic solv problem explicitli control avatar move user avatar imit motion share classic problem video user bodi resid space radic avatar flaw appar multipl user gaze pattern orient inform gather user monitor map appropri avatar stand group avatar track lend desktop virtual environ approach avatar design adopt explicit treat avatar autonom agent act accord world inhabit similar avatar autonomi limit rang commun express face leav user direct control navig speech content avatar show behavior base current situat user input control higher level current system approach start address control user manipul repres user current intent respect aspect anim human figur spontan avatar show spontan involuntari reaction user initi explicitli discret user avatar updat carri behavior synchron gap occurr user input lag time bridg produc seamless anim map user space user user avatar resid drastic environ direct map project live imag user avatar produc avatar action control intent level autonomi level involuntari commun behavior avatar give cue virtual situat human commun behavior order autom commun behavior understand basic mechan human human commun activ particip effortless synchron particip occur natur facilit number channel modal convei inform partner channel includ word inton hand facial bodi orient ey gaze give feedback avoid overlap partner give feedback secondari facial receiv inform speech channel channel work supplement complement salient point direct listen attent provid addit inform elabor multipl channel emploi refer multimod current work gaze commun facial express fundament establish maintain live link particip gestur bodi postur requir elabor articul human bodi scope current work illustr meant commun section describ scenario peopl meet behavior emploi referenc background studi relev page number includ paul stand interest peopl susan walk mutual glanc paul nod susan paul smile susan touch shirt dip smile approach paul back paul initi close meet smile paul tilt head side slightli offer susan shake lightli face repli step side face paul angl start paul susan displai gaze start long utter mark syntact event speech facial eyebrow question nod eyebrow word give feedback listen form low eyebrow action final give floor person gaze speaker partner eyebrow emphasi person give floor attend clue thrown absenc implement commun behavior function volit statu distinguish user speak end bodi meet gaze chosen relat work embodi distribut virtual environ research issu system massiv crg nottingham techniqu design issu made clear involuntari facial express gestur import hard captur avatar autonomi suggest popular internet base chat system connect number user graphic earli world shown graphic user compel altern pure system system natur integr graphic commun place studi human commun behavior seldom consid design believ avatar work includ judith particip gaze direct displai microsoft comic chat illustr imag automat gener interact user lack bodi system implement function head comic broken discret exclud possibl thing feedback subtl gaze behavior creat fulli autonom agent capabl natur interact entail integr gestur facial express appli knowledg analysi studi social develop system anim anim render graphic autonom agent engag system dialogu planner gener accompani commun base agent initi goal knowledg autonom agent capabl carri user emploi rang commun behavior manag flow system good exampl theori studi human commun appli comput concern user embodi issu avatar control anim lifelik humanoid figur greatli improv recent year improv system demonstr visual appeal humanoid anim tool script complex ideal agent avatar similarli humanoid project deal virtual actor perform script behavior automat gener commun behavior synchron actual user address system extern control anim autonom actor call method direct anim behavior number level aliv oz project goal set behavior focu behavior accompani languag time distinct phenomena commun behavior system demonstr autom commun behavior avatar system consist client program server program client render singl user view distribut virtual environ figur user connect server avatar model repres upper bodi humanoid charact user navig avatar cursor give command paramet avatar interact textual user chat window user choic avatar commun behavior reflect user current intent avatar knowledg commun rule user intent set control paramet user client connect produc behavior user remot avatar implement control paramet tabl potenti partner user interest user potenti partner click avatar visibl view window anim visual cue chosen avatar turn react user avail avail user welcom peopl show interest effect initi exchang glanc perform confirm newcom partner chang avail effect switch toggl switch control panel figur user break user inform system intent break place special symbol forward chat string elicit divert give partner visual cue word spoken readi leav paul type back partner paul avatar glanc displai word partner repli break broken mutual partner repli normal break cancel continu partner produc subsequ break avatar move broken gener behavior discuss commun essenti make clear distinct phenomena hand commun behavior phenomena describ intern state user refer event phenomenon phenomenon set commun reveal state peopl phenomenon head wave smile behavior avatar react event select phenomenon describ initi execut paramet type descript potenti partner avatar id person user chat avail boolean show user chat break boolean show user stop chat tabl control paramet reflect user intent commun behavior essenti avatar behavior control consist flow execut top bottom figur reaction event tier defin entri point behavior control tier implement set function call client messag arriv network avatar environ updat function list tabl tier heart avatar decid react situat reaction involv pick phenomenon describ state avatar pick situat close user current intent figur avatar behavior control consist flow execut top bottom reaction event phenomenon commun behavior commun behavior commun behavior anim geometri anim geometri anim geometri user environ function event user move avatar partner move avatar reaction rang avatar rang user show interest avatar show interest partner end avatar spoke user transmit utter word displai user avatar word utter displai tabl behavior control function implement reaction event phenomena tier implement map state select event set visual behavior tabl map base previou work human commun commun behavior start anim engin manipul avatar geometri order chang visual sampl interact overview section describ typic session illustr imag show express avatar imag present sequenc snapshot reflect chang time interest user scout seek interest chat spot lone figur appar occupi click phenomena commun behavior approach react smile smile embrac smile plan wave tabl map phenomena visibl behavior potenti partner avatar react glanc chang express lack sustain attent signal user autom sequenc glanc shown figur figur sequenc glanc user click avatar express chat user partner found user continu scout person chat notic lone figur decid repeat attempt time express receiv invit indic user autom sequenc glanc figur figur sequenc glanc user click avatar express chat user immedi express mutual avatar automat exchang distanc confirm system consid partner close automat exchang bs rang figur show sequenc figur avatar exchang distanc system regist partner close exchang exchang start sentenc broken word displai head avatar word avatar accompani express anim utter figur figur word accompani special facial express nod mark elicit eyebrow end utter chat produc break utter place forward slash begin sentenc make avatar divert gaze word shown figur user notic behavior decid respond end avatar automat wave break ey contact figur user mark sentenc break avatar displai divert gaze word give subtl cue partner evalu present approach avatar mere visual integr part sheer mere embodi interact user choic autonom commun behavior user concentr high level control depend avatar convei commun signal repres user commun intent approach potenti limit appar concern fact commun behavior adher gener fulli understood comput model simplist constrain behavior limit set displai real world nuanc concern system capabl natur human commun concern relat balanc autonomi direct user control wider field research promot direct manipul claim user strong desir gain system promot treat system agent carri instruct talent current avatar accur reflect user reliabl undermin user left uncomfort skeptic state trend distribut virtual environ anim avatar kind anim detract commun power system avatar simpli repres sheer mere methodolog order respond carri evalu concentr dimens solicit subject system experienc graphic chat told goal meet user find user told interact briefli quit time distribut virtual environ color plai script person life avatar respond attempt enter order test function avatar lightli mirror subject similar verbal devic user commun kind produc subject interact version subject version ident autonom version avatar capabl commun behavior gener function user intent avatar manual version avatar capabl behavior autonom behavior gener menu menu select list nod toss shake glanc glanc smile eyebrow avatar gener autonom commun user gener addit behavior menu ident menu manual condit possibl compar autonom condit version avatar capabl navig produc commun behavior move subject taught system sheet list behavior avatar capabl short interact ensur function system told avatar peopl space ident function end subject told fill questionnair evalu natur experi item natur commun power system question natur avatar behavior system success system support rich user control system control perform avatar encount good person express subject request list fact acquir put user item serv measur task perform questionnair item measur point scale result set concentr user percept natur autonom judgement commun power judgement control autonom manual version system final set compar avatar embodi commun behavior capabl gener behavior gener descript autonom condit significantli longer manual condit index interest subject pursu interact autonom system subject autonom condit rememb fact peopl interact subject manual condit condit index engag subject attent divid particip control avatar address natur user natur avatar behavior natur interact subsequ question show user autonom system judg natur user condit figur order address commun power user feel understand peopl feel express peopl peopl met understood understood meant feel user express anova subsequ aggreg data reveal user autonom system judg express user manual significantli express condit autonom commun behavior express autonom gener commun behavior behavior user avatar kind behavior found experi tediou user autonom system order address issu control user control control natur express control manual auto figur effect manual autonom avatar interfac percept natur user subsequ question reveal user autonom system consid user control user manual system nonverb commun behavior gener user freed worri felt control point autonom gener embodi action avatar recal autonom commun behavior gener function user word punctuat choic emphasi head nod gener user type eyebrow user end sentenc proxim avatar argu autonom system manual behavior chosen keyword menu choic order test keyword examin condit categori keyword demonstr total keyword differ condit subject languag condit seemingli normal keyword kind control languag turn comparison autonom condit condit design suggest simpl exhaust avatar featur simpli distract user commun final autonom condit shed light issu autonom condit judg significantli natur condit condit judgment express judgment user control subject spent time condit manual condit time spent autonom condit find support drawn user deriv control pleasur primarili textual contribut feel distract anim avatar commun behavior user feel chat space equal express avatar embodi commun behavior avatar autonom embodi commun user concentr concentr instanti text text anim commun behavior lead question shouldnt build system avatar simpli repres primari issu set address design avatar system resolv add behavior simpl issu deserv discuss posit goal graphic chat system make interact human natur user judg autonom system natur system commun behavior context autonom commun behavior greatest effect avatar recal user unsur speak user test problem worthi investig caveat mind interpret result evalu directli test user prefer system user engag system direct comparison difficult pilot show short interact system left user unabl direct comparison requir period minut minut long period time introduc unwant fatigu effect introduc unwant comfort factor subject judg system simpli time comfort manual version system similar system user prefer manual system autonom manual system current user judg manual system highli scale fun experi simpli judg autonom system highli paper introduc approach design implement draw literatur context autonom commun agent argu todai avatar serv presenc contribut experi order understand import commun function reli previou research commun human research develop system emploi find autom commun behavior avatar system stand pass repertoir commun begin essenti cue initi import continu ad model draw psycholinguist literatur real world empir studi conduct domain mind examin concentr issu floor manag work physic model arm hand line autonom gener hand gestur text speech term ongo discuss field direct manipul autonom introduc balanc support view promot direct manipul claim user control master interfac interfac action stai close task domain minim mental command make perfect system user carri spare burden execut involv depend virtual set current fulli determinist type execut fit job descript job treat user action instruct mean differ base context context accomplish rich involuntari behavior social reli explicit user control exploit function embodi construct anim avatar avatar person agent user capabl natur initi sustain valuabl contribut research broader theori embodi agent figur user avatar acknowledg member gestur narr languag group media lab commun valuabl discuss research support part nation scienc foundat award media lab digit life thing note item adapt questionnair request author tabl mean addit statist request author refer casei park social virtual realiti system spoken runtim technic report gaze mutual gaze cambridg function ws sigart embodi collabor virtual direct autonom creatur realtim virtual siggraph role gaze initi social integr face gestur automat spoken dialogu spoken dialogu system mit gener facial gestur spoken inton multipl siggraph power nod envelop emot feedback anim journal appli artifici press prevost model multimod wilk machin displai research languag social illustr multimedia tool resourc organ mutual conduct pattern behavior encount cambridg press negoti context goodwin rethink languag interact phenomenon cambridg press siggraph softwar ibm system manipul interfac number acm hand gestur reveal thought comput psycholog person type commun system script interact actor virtual siggraph note contrast gener synthesi spoken icslp american embodi humanoid capabl multimod dialogu actor perform script 
Agents/781 1 multiag system specif uml statechart aim intellig manufactur frieder stolzenburg informatik universit institut ur rheinau koblenz multiag system specif uml statechart aim intellig manufactur arai frieder stolzenburg knowledg industri dept institut ur informatik advanc system center fachbereich informatik mitsubishi materi corpor universit rheinau koblenz tokyo germani japan abstract multiag system paradigm contribut field theori technolog develop order design multiag standard procedur present industri applic complex structur plenti work resourc requir standard specif method standard method design softwar kei word simplic wide accept method multiag uml statechart aspect multiag advantag type diagram appli method robot soccer network applic approach enabl standard design multiag automat translat specif run implement verif formal analysi rigidli formal manner system specif concentr paper emerg author visit octob septemb formal specif multiag system gener applic robot network applic homogen heterogen gener approach softwar methodolog standard agent multiag unifi model languag introduct softwar system view multiag system provid good mean robust architectur softwar system system part function multiag system interest industri applic research foremost research topic field artifici intellig present implement system systemat formal clean show appli robot soccer activ industri applic activ web servic applic draw feasibl enhanc creat nowadai internetwork technolog import context respect multiag wide accept method design network enabl flexibl robust system industri applic advanc manufactur design multiag system overview applic agent technolog motiv multiag system system intellig agent pursu set goal task agent comput entiti softwar program robot view perceiv act environ autonom behavior partial depend experi import system agent design implement current softwar specif techniqu adapt respect multiag coordin agent express method make type diagram specif multiag uml statechart advantag procedur hand design programm rapidli develop hand specif easili turn execut type diagram formal base statechart verif formal analysi multiag model check focu laid rigidli formal specif method multiag basi formal system demonstr applic approach applic robot soccer detail featur uml statechart exploit develop multiag system translat specif execut interest reader refer overview paper distinguish system homogen agent principl system heterogen agent abil perform task team soccer agent view system homogen web servic peer peer applic system heterogen agent applic simul robot soccer section network applic section show kind multiag system introduc section multiag system studi domain robot behavior agent includ collabor mean uml statechart method applic industri introduc network applic domain order investig effect method domain discuss gener problem multiag system mean uml statechart relat work state machin dynam system appropri state machin statechart specif dynam procedur aspect softwar industri applic order provid understand give formal definit section demonstr statechart good mean dynam system multiag independ agent interact basic compon statechart part uml accept mean dynam behavior softwar system rigor formal flexibl implement analysi multiag system state connect transit condit action annot repres rectangl round corner structur hierarch transit drawn direct arc interconnect state transit annot event recept request chang guard action state composit behavior agent state sequenc simpl state finit configur reader refer section state machin figur sketch behavior robot soccer agent discuss detail section detail hidden decomposit icon machin simpl state init composit state concurr state commun start state initi state init state attack defend belong main state behav initi state init permit transit annot empti guard correspond agent center point agent kick ball full power enter attack state attack defend state mutual connect transit agent ball attack defend state transit attack defend back score goal attack mark commun defend init behav figur state machin behavior configur step behavior state machin introduc concept configur step definit configur root tree root node topmost initi state state machin configur complet leaf node label composit state initi state introduc successor leaf node label concurr state tree branch point state machin model behavior agent act environ main effect agent interact environ state machin chang situat form sequenc situat situat defin map variabl valu current world includ agent configur agent perform step configur proper multiag account agent compon agent perform step time distinguish part figur show configur state machin transit init attack reader refer briefli summar definit definit configur state machin configur mean transit state state annot current written node label ident replac complet annot event guard hold actual situat revisit figur transit init attack annot state step configur shown figur definit init node replac attack attack concurr state commun belong successor node attack state complet triangl symbol behav init behav commun attack figur configur transit definit configur configur simultan state occur path configur tree obtain appli model concurr behavior descript multiag entiti interact concurr heavili make concurr state applic section concurr state prerequisit proper execut parallel applic introduc applic robot soccer network bit detail robot soccer section tackl author consid network applic section interest thing applic appli techniqu describ multiag system section continu section address problem interact possibl heterogen agent robot soccer robocup soccer game chosen central topic aim innov appli social problem industri robocup intern joint project promot artifici relat field order perform soccer game robot technolog incorpor includ design principl autonom multiag strategi acquisit reason robocup consist leagu real robot size simul robot simul leagu offer softwar platform research softwar aspect robocup softwar design aspect import soccer agent design approach figur mode level abstract desir agent script level plan skeleton long mode chang role skill level host basic action soccer agent ident intern structur agent homogen natur state behavior agent state machin heterogen agent system multiag system statechart section guard logic good idea implement state machin logic program languag prolog approach develop soccer agent declar present advantag make prolog guard condit express yield specif process program conveni implement tree agent programm full program languag design implement agent restrict soccer domain homogen agent agent state machin agent role depend respect situat doubl pass situat interact agent trigger synchron state conveni skill skill skill skill figur layer architectur statechart hierarch structur make modular specif multiag system assum soccer plai agent modul move kick control independ order perform agent aim oppon goal kick devic time ball shot initi posit agent observ action requir synchron move modul kick modul shown figur synchron concurr region state synch state introduc uml synch drawn conjunct fork join shown thick bar figur left bar fork bar join state fire outgo transit synch state limit bound differ number time outgo incom transit fire bound set ball kick simul synchron introduc variabl rang synch initi effect transit synch state express set condit action transit fork join state shown figur increment introduct make synch state superflu note addit annot treatment synchron shown transit annot shown pass adjust goto shoot aim move kick figur machin concurr state pass network applic internet technolog widespread industri plenti data exchang internet data includ html page numer plant oper data exchang internet involv problem commun protocol internetwork technolog applic technolog approach design scheme data exchang multiag system method robot soccer introduc network applic domain multiag system concurr action collabor agent explicitli recent internet network gener prevail accompani advanc induc develop industri applic integr manufactur resourc achiev mean applic base multiag system network applic expect plai import role real intellig manufactur compon facil distribut show industri multiag system design method robot technolog investig design principl autonom multiag strategi design multiag system shown figur heterogen user broker proxi agent sensor behavior user agent commun send receiv file support broker agent agent primari function invok extern event invok user broker agent alloc destin file user agent broker agent enabl function broker agent user agent directli commun proxi agent introduc sensor agent network condit notifi result user agent optim user agent behavior heterogen agent function multiag system mean statechart case action execut parallel express region statechart detail section multiag system show behavior singl agent mean domain robot soccer network applic featur multiag system repres method remark robot soccer network applic method feasibl import aspect multiag domain characterist function approach enabl design multiag verif formal analysi system strict formal long run switch role shown figur user agent design state behavior similar soccer player agent figur toplevel behav user agent put state correspond mode level script level approach robot soccer domain figur exploit hierarch structur statechart figur architectur network applic figur state machin user agent extern user agent provid put assum user agent put analog attack defend state robot soccer domain put user agent transfer file state agent receiv file action initi click event discuss put state greater detail figur synchron put state composit state concurr send depict figur send design provid file transfer introduc order ensur function synchron concurr state enabl synch state render circl concurr region statechart synch state fork join thick bar stand transit concurr state synch state enabl synchron consequ figur synchron agent collabor statechart repres collabor collabor heterogen agent figur illustr concurr rightmost region belong sensor user agent figur design perform check task transfer substat kind mirror order enabl fast file transfer synch state repres collabor agent concurr singl agent interact agent parallel thread activ agent express synch state state machin synch state translat simpler synch introduc dedic synchron variabl figur transform concurr region execut possibl account definit action agent coordin extern event variabl synchron realiz state code explicitli prolog relat work method specif multiag system make standard softwar engin method formal easili achiev execut formal system analysi verif combin featur origin discuss work extend unifi model languag agent work analysi agent system case overview applic multiag process inform electron medic reader refer agent design extens uml object manag develop uml extend uml agent bunch diagram express aspect multiag templat packag sequenc collabor state activ diagram import aspect system agent interact collabor figur collabor agent stress interact protocol detail level combin sequenc state diagram good idea put nest present interest statechart label agent perform respect action type direct transform specif contrast approach present softwar engin describ system level abstract agent engin introduc level agent level previou approach discuss exploit uml design multiag system make standard diagram languag class diagram ontolog commun languag fipa acl kqml addit type architectur diagram set agent class relat protocol role diagram introduc relat interact protocol uml idea make standard notat author design multiag system mean diagram type agent design analysi approach composit verif multiag system agent hierarch standard notat statechart class diagram approach tempor logic composit verif abstract level base agent system consist number agent extern world formal descript exist multiag system approach nice theoret point view tool approach softwar engin verif specif standard procedur graphic notat similar uml collabor diagram base procedur multiag system emploi tool technolog develop model check class logic obtain tempor logic tree logic modal logic system sever agent multiag finit state machin modular increment design analysi multiag system standard softwar engin remedi approach approach show specif multiag system statechart formal analysi model check design analysi techniqu import industri accept approach combin softwar engin artifici intellig aspect major emphasi paper final remark present method specif multiag system appli robot soccer homogen agent network applic system heterogen agent present paper advantag approach present multiag system type statechart translat automat execut prolog commun agent statechart explicit desir subject work work concentr refin formal specif multiag system automat analysi system properti interest formal verifi multiag specif translat run code program languag approach present lai basi formal specif verif multiag system refer serafini model check algorithm system intern workshop intellig agent agent lnai page poggi exploit uml design system engin societi agent lnc page treur composit verif system tempor logic journal languag grefenstett strategi acquisit genet algorithm handbook genet chapter page van nostrand wooldridg roadmap agent research develop autonom agent murrai soccer agent uml fachbereich universit murrai agent uml robot soccer stolzenburg logic approach soccer agent engin robocup robot soccer world cup lnai page object manag model languag septemb version bauer extend uml agent agentori inform system workshop nation confer artifici page pnueli semant statechart ito intern confer theoret aspect comput lnc page york stolzenburg reason cognit robot system und de spatial agent implement logic express languag robot soccer lnai page editor multiag system modern approach distribut artifici intellig mit research report frieder stolzenburg multiag system specif uml statechart aim intellig manufactur kurt lautenbach reproduc empti mark jan murrai agent uml robot soccer andrea winter exchang graph anna simon slice book technolog ein technik fur ein bernt volker approach enabl program understand preprocess languag frieder stolzenburg specif multiag system statechart formal analysi model check oliv ration agent statechart util function torsten urgen ebert conceptu model web site gener graph technolog carlo che urgen frieder relat normal logic program transform properti harri andrea winter appli gupro geo case studi pascal von stephan model concurr algorithm urgen bernt franz lehner workshop softwar reengin mai stephan workshop und fur jan oliv frieder stolzenburg logic approach soccer agent engin peter zhang ftp intern workshop theorem st juli frieder garc carlo che introduc gener specif logic program manfr specif symbol implement constraint peter fabio tame richard andrea andi sch standard exchang format urgen roger formal ulrich fuzzi techniqu cach memori jen implement urgen bernt franz lehner workshop mai peter michael abduct model construct urgen bernt andrea winter interchang format reengin tool frieder oliv jan bj spatial agent implement logic express languag kurt carlo simon zur hybrid system frieder stolzenburg power model gener peter jd bruce spencer merg path improv minim model hyper tableaux urgen roger uh worldwid peter norbert ulrich furbach confluent connect calculu bernt andrea winter queri enabl technolog softwar reengin urgen georg pick program urgen ulrich ilkka niemel nonmonoton effici calculi implement urgen steffen infer mechan theori applic ws ki jose urgen mauricio claudia semant logic program stefan urgen teodor przymusinski super logic program urgen dix logic program paradigm stefan urgen burkhard ulrich zukowski comput model manfr kamp greql ein fur da peter urgen angelika manfr andrea winter und formal grundlagen peter da org thoma marx workflow model wfmc dieter obel schedul criteria ag constraint process hard system ulrich furbach disjunct logic program horn program control program andrea schmid solut count infin problem distanc vector rout ulrich michael frieder stolzenburg proof debug peter sch model elimin simplif applic softwar verif bernt andrea peter urgen ebert program system urgen jorg lobo logic program nonmonoton reason kurt carlo jan technischen anwendungen manfr kamp manag softwar repositori program tool gener approach 
HCI/2352 4 architectur implement java packag multipl input devic juan pablo benjamin bederson interact institut advanc comput studi comput scienc depart colleg md usa abstract major difficulti write singl displai groupwar applic input multipl devic introduc java packag address problem offer architectur access advanc event java describ architectur limit mid briefli describ applic mid input multipl keyword singl displai groupwar cooper work multipl input devic window serial bu introduct coordin brought peopl desktop groupwar applic lotu note microsoft lead commerci product field cooper work peopl collabor place time comput peopl collabor place time singl displai groupwar model support collabor work sdg applic make user collabor singl comput displai multipl input devic sdg applic deal problem rule interact consid rang issu builder sdg applic provid interact figur children mice comput run applic wrote mid metaphor make sdg challeng applic user issu incompat sdg applic deal global state focu user set tool screen space critic question address focu technic issu input multipl devic comput consist mechan introduc multipl input devic java packag support input multipl devic singl comput develop java abil write sdg applic power straightforward input multipl devic gener architectur kind input devic chose java platform independ sdg applic run multipl platform mid open sourc softwar gnu librari public mid consist java nativ layer layer extend devic type nativ class implement devic platform java layer extend standard java event class applic mid chang platform mid support serial bu mice window pick combin start wide avail instal gener devic window nt support microsoft inform version nt input multipl mice window issu version nt registri kei set nt merg input multipl mice mid chose mice devic support good percentag person comput user familiar chose standard devic connect singl port grow standard expect domin protocol input devic futur mid written larg java design fundament tradeoff made portabl featur set mid offer java support event decid goal mid enabl applic access input devic capabl work manner work applic mid chang run underli implement mid updat devic platform natur access special input devic mean input devic physic applic deal accommod user input devic good applic mid work mid nativ code unavail mid revert offer standard java event relat work peopl problem input multipl devic stewart solv problem window solut support serial mice design mmm earli sdg environ support input mice pebbl project investig person digit assist input devic sdg applic pda input devic fine choic peopl special prohibit peopl input devic control approxim time cost standard mid input pda mice advantag pda young dont lot control input devic studi effect protocol children learn collabor environ mice connect simultan game input devic investig multipl user collabor singl comput control aspect input rekimoto comput pda special architectur support comput receiv input multipl focu research sdg requir special comput connect type pick drop video game system input multipl devic support limit sdg applic system part support joystick easi program buxton investig person hand simultan interact comput area implement custom code receiv input extra typic tablet serial port mid mid extend standard java event mechan order applic replac standard java event extend mid event current support window applic mid access multipl mice provid mid dynam link librari present system refer system run window mid dll mice system applic mid work applic mid run system nativ dll file instal mice present oper system window mid revert java event applic mid advantag input multipl mice run readi work run system mid event versu java event mid similar java event mid extend java class event applic typic event event event java event class receiv java event implement interfac order regist compon visibl window event event call method compon class implement method interfac method call event object describ event pass top figur show sampl code class java event mid event bottom figur show sampl code class mid event notic differ code code java event extra line code mid sourc object compon pass mid sourc object nativ code handl visibl window order work regist class regist mid sourc object listen mid event mid sourc object gener mid event mid offer flexibl regist listen devic devic support applic written style style dispatch event devic listen listen determin devic gener event call return id gener event altern applic style regist listen receiv event specif devic applic write listen support decoupl devic call return number mice turn applic build intern data structur mechan support input devic differ code interfac registr event prepend mid sourc object support extra featur develop set locat motion time span count click set locat mice applic defin cursor give initi locat screen constrain motion mice motion mice cursor visibl area window applic divid screen region oper region import extra featur mid enabl applic access input button movement ax wheel gener button event motion axi mid architectur section describ design mid figur show mid class structur mid nativ code mid work system provid develop mid sourc class singleton implement method return mid multi sourc mid singl sourc mid multi sourc object return run system nativ code input multipl mice mid singl sourc object return run system listen java event turn mid event mid system mid multi sourc class heart mid system care regist check creat mid broadcast listen care provid extra function develop mid sourc object system order track listen mid multicast mid nativ code gener event microsoft api receiv event mice assum keyboard order support event access keyboard alt modifi kei access keyboard event demonstr low level assumpt singl keyboard made mid support multipl deal differ java code standard java event access motion event class implement constructor add motion listen public listen event public void public void motion listen event public void public void figur top code fragment show java code standard java motion event bottom code fragment show java code mid event access multipl mice java code mid event access motion event multipl mice class implement int number mice connect system public listen event public void int devic id manner public void int devic id manner motion listen event public void int devic id manner public void int devic id manner java event queue event coalesc elimin redund event event post motion event replac exist motion event queue write custom version event coalesc mid event make event mice coalesc mid system mid singl sourc object run system wrap standard java event regist pass event applic fact mid singl sourc pure java code run platform support java figur summar mid featur extend mid mid extend ad support multipl mice ad support devic ad support multipl mice platform consist creat nativ librari input multipl mice interfac correctli mid multi sourc class librari implement nativ method declar mid call mid event occur ad support devic complex devic type event listen interfac adapt mid mid sourc class modifi limit limit aspect mid input multipl mice window mid dll mice mice mice mice limit window window merg input input mice channel limit mid mid multi sourc sourc mid mid system cursor user send input figur class structur mid uml solid line solid arrow repres inherit dash line solid arrow repres implement solid line open arrow repres associ number text associ line repres cardin role class associ throw throw throw catch creat creat applic switch applic keyboard altern mice control singl system cursor perform gener mid event notifi mid suffer minim overhead mid multi sourc sourc mid event mid event gener mid singl overhead java event pass mid event listen applic listen sampl incorpor mid applic user interfac tailor sdg author tool children includ type hyperlink tool zoom capabl support collabor creation stori children singl comput simultan child interact softwar previou version written perl librari support zoom recent java java packag support graphic zoom perl version input multipl mice linux comput run version tightli ti unix make port platform difficult problem target user access linux success sdg applic good part reason success design user interfac local tool menu tool palett local tool graphic object sit data surfac data local tool pick input devic approach natur suit sdg global state pen color tool state local tool act cursor crucial current oper system support multipl cursor aim rewrit java make multipl advantag replac earlier zoomabl graphic packag local tool java version facilit make sdg applic dai work modifi standard java event code input multipl mice mid run system java combin java nativ code multipl input multipl mice design team concentr figur interact rule user share workspac futur work input plan add support plan make fulli function mid platform focu unix system run window mid possibl provid input event applic nativ system gener event recogn voic data applic written access voic data mid event mechan consist input devic hope mid uniform architectur support multimod input feel mid make easier build sdg applic remov pain problem input multipl devic fact applic mid run platform support java ad advantag mid similar java develop java experi find easi incorpor mid applic room mid step direct acknowledg work motiv earlier work jason stewart mexico allison druin mexico group experi input devic code final notion gener architectur support sdg jon meyer share experi write java code tablet figur summari mid featur java api nativ code devic platform applic awar detail nativ code defin event input mechan support mice window applic continu work nativ code featur reduc standard java featur extra wheel access input devic featur wheel extend support input devic platform earli experi iyer write earli version nativ code understand microsoft work version mid support part collabor project partner swedish institut comput scienc royal institut technolog stockholm nottingham england fund european union school initi current version singl user elementari school stockholm grate children feedback design idea eager learn design sdg applic bob darpa instrument support wouldnt exist wouldnt exist support refer local altern tool palett user interfac softwar technolog acm pp zoomabl graphic sketchpad explor altern interfac physic journal visual languag extens graphic toolkit java user interfac softwar technolog acm user interfac architectur share editor singl screen user interfac softwar technolog acm pp support cooper control object multimedia applic extend abstract human factor comput system acm pp studi input human factor comput system acm pp design children technolog san morgan kaufmann design collabor educ human factor comput system acm pp design element reusabl objectori softwar effect protocol children learn collabor environ graphic interfac canadian inform process pp design gui paradigm base transpar human factor comput system acm pp tablet human factor comput system acm pp collabor multipl pda connect pc comput support collabor work acm direct manipul techniqu multipl comput environ user interfac softwar technolog acm pp singl displai groupwar doctor nm singl displai model collabor human factor comput system acm hand enhanc collabor singl displai groupwar extend abstract human factor comput system acm pp 
HCI/2029 4 present futur research ubiquit comput gregori abowd elizabeth mynatt colleg comput gvu center georgia tech ga abstract comput physic world ubiquit avail comput suggest paradigm interact inspir constant access inform comput capabl past research ubicomp push interact natur autom captur access chart futur research ubiquit review accomplish effort point remain research challeng research ubiquit comput implicitli requir address notion number type physic space distribut comput number peopl system posit area applic everydai focuss scale interact respect time push avail comput tradit desktop fundament relationship human provid continu interact move comput local tool constant companion inher interact theme difficult issu social implic ubiquit comput challeng evalu ubiquit comput research cumul experi point lesson visibl simpl guidelin steer research effort akin effort involv evalu strategi form spectrum technolog feasibl effort studi perspect ubiquit ser natur captur everydai augment social implic introduct weiser introduc area ubiquit comput put vision peopl environ augment comput resourc provid inform servic desir past ubicomp research attempt augment implicit goal assist everydai life overwhelm vision devic vari rang size person devic share devic devic commonli devic person digit digit electron whiteboard develop infrastructur support continu mobil comput arriv aspect vision applic emerg leverag devic infrastructur ubicomp suggest paradigm interact inspir widespread access inform comput capabl explor applic perspect evolv decad start ubiquit comput project xerox parc review accomplish outlin remain challeng desir natur interfac facilit richer varieti commun capabl human comput goal natur interfac support common form human express leverag implicit action world ubicomp applic adapt behavior base inform physic comput environ larg number ubicomp applic strive autom captur live experi provid flexibl access experi undertak issu scale implicit definit ubicomp research weiser defin notion scale broad space comput devic scale system respect distribut comput physic space reinforc desir break human interact requir critic mass accept collabor impli scale respect peopl final present challeng scale system push avail interact basi uncov class larg unexplor interact push ubicomp research centuri address scale respect section introduc call everydai promot inform unstructur activ typic everydai live activ continu constant flow action clear start end point familiar exampl orchestr commun famili manag inform structur paper evolutionari path past work ubicomp emerg step demonstr parctab comput form factor comput applianc push tradit area comput scienc network oper system form factor work tradit input devic keyboard develop input hour dai week capabl step work shorthand languag parctab initi infrastructur deploi devic gener numer tour guid system mimic activ badg built deploi real point applic design begin work system develop implicit user input minim technolog everydai life object research understand everydai task alter introduct ubiquit technolog ubicomp applic support common meet task parc tivoli result wai organ materi meet captur environ educ set provid opportun understand pattern longer term review task larg multimedia record applic wearabl comput initi constant access tradit individu access recent applic attempt augment memori provid implicit inform share group direct applic weiser deem ultim ubicomp deepli influenc authent extend ubicomp system todai start understand implic continu comput futur hold constant avail tool assist task wear comput abil comput alter percept physic support constant connect distant peopl provid inform continu partner thought action offer offer killer exist overview investig histori ubiquit comput explor interact autom captur access live experi addit accomplish applic research outlin remain research challeng hci research pursu millennium explain necess ubicomp research explor continu everydai activ area research motiv applic build earlier theme move ubicomp realm everydai comput continu interact inher interact theme difficult issu social implic ubiquit comput challeng evalu ubiquit comput research conclud reflect issu case current strategi evalu ubicomp system comput natur interfac ubiquit comput inspir applic develop implicit assumpt physic interact human comput current desktop paradigm human interact physic world human gestur write commun human alter physic artifact natur action explicit implicit input ubicomp system comput interfac support natur human form commun begin supplement replac element gui interact paradigm interfac learnabl gener abil support task author draw drastic chang structur task peopl disabl tradit keyboard access work year emerg area perceptu interfac driven research commun comput vision comput percept interact realiz failur gener pen comput research suggest techniqu object physic world manipul electron creat graspabl tangibl interfac harrison al attach sensor comput devic order provid wai physic manipul devic interpret appropri applic run devic applic support natur interfac leverag input output modal attempt review impress amount work natur focu issu import enabl rapid develop effect natur interfac import area discuss multimod theme confer journal natur data type develop applic natur handl form input easili keyboard input raw data signal underli natur sensor type interact system develop expect user interfac toolkit develop basic level support oper textual primit keyboard interact basic support manipul speech provid speaker cue select speech segment speaker identif video ink physic devic manipul detect sensor interact interest comput focuss recognit techniqu convert pen input text person requir ink text user convert handwrit form effort put standard support pen input format exchang pen input platform effort defin effect mechan manipul freeform ink data type program kind oper support natur data type tivoli system provid basic support creat ink data distinguish freeform ink data implicitli structur gestur moran featur freeform ink abil merg independ stroke form word segment languag produc note classroom detail tempor spatial heurist merg assign timestamp chiu wilcox produc gener dynam base hierarch select link audio ink structur techniqu standard applic develop creat interfac work chiu wilcox structur techniqu appli natur data type interact interact natur interfac set problem permit kind mistak recognit error initi reaction system design elimin improv recognit accuraci reduct absolut error rate major peopl notic differ speech recognit system elimin error human make mistak deal form commun handwrit recognit expert handwrit recogn recognit accuraci low human accuraci cursiv handwrit print handwrit perfect evid suggest hat comput handwrit recognit perfect recogn human data human process power variabl fatigu usag data differ significantli train reduc recognit accuraci time recognit accuraci determin user satisfact complex error recoveri dialogu benefit affect user satisfact found user frustrat recognit error task enter command form write journal entri suggest enter singl word case command larger compar effort enter word paragraph journal entri error handl problem design comput system attempt mimic human abil research area error handl recognit technolog assum error answer question wai deal research area error handl interfac error reduct involv research improv recognit technolog order elimin reduc error focu extens easili subject paper evid suggest holi elimin achiev error discoveri system user action relat error occur system told error explicit user user find error effect output uncertain interpret recogn input techniqu autom error discoveri threshold confid histor statist explicit rule specif reusabl infrastructur error correct toolkit provid reusabl compon class similar problem exist interfac error handl benefit tremend toolkit present librari techniqu input toolkit handl inher ambigu multipl interpret gener raw input prototyp toolkit al support reusabl recoveri challeng remain comput compel earli demonstr ubicomp olivetti research activ badg xerox parctab applianc devic leverag simpl piec user provid valuabl servic call forward phone automat updat map user locat connect comput devic physic world control system autonom guid satellit missil exampl simpl applianc demonstr link implicit human activ comput servic serv augment gener human activ locat common piec context applic develop widespread applic car navig system handheld system vari content displai unit physic locat exhibit area abowd cheverst import piec context recogn individu object earlier system recogn sort barcod identifi tag recent work includ base recognit al demonstr devic wall piec machineri rekimoto figur recogn color barcod overlai addit inform object video displai recent effort attempt substitut visual object recognit strategi object individu tag numer system leverag ident locat system difficult implement salber al creat simplifi implement evolv applic work strict separ context storag reaction contextu separ facilit construct applic al point common design challeng creat believ experi interfac note interfac kei person associ addit displai movement physic world figur lk figur rekimoto system handheld devic recogn tag object overlai inform scratch surfac comput issu address discuss challeng incorpor context repres ubiquit access context context coupl context natur interact provid effect augment realiti context posit ident system incorpor knowledg histori long peopl piec inform environ complet definit context context good minim set current system focu interact ident user rare incorpor ident inform peopl environ human tailor activ recal event past base presenc peopl interact current system assum user leav question open perceiv interpret human activ difficult problem interact continu devic incorpor interpret human activ provid inform compon context explor coupl notion contextu tour guid system learn histori movement physic tailor inform displai base perceiv path interest user idea fuller explor except time index captur record summar long person applic unawar passag time interest understand rel time aid interpret human activ visit exhibit indic gener lack interest baselin behavior action violat perceiv pattern interest home notic elderli person deviat typic activ morn routin challeng perceiv person understand form contextu inform give indic affect state bodi skin place start context relat definit context question repres context good applic develop left develop limit scheme store manipul kei inform evolut sophist enabl wider rang capabl true separ context programm reaction context ubiqu context context fusion obviou challeng comput make ubiquit posit shown singl sourc context servic posit good gp work suspect urban region varieti indoor posit scheme differ characterist term requir singl solut meet requir solut obtain ubiquit context assembl context inform combin relat context servic context fusion similar intent area sensor fusion context fusion handl seamlessli hand boundari context servic negoti resolut strategi integr inform compet context servic piec context concurr provid servic fusion requir technolog reliabl determinist combin measur multipl sourc confid interpret context fusion assist provid reliabl ubiquit context combin servic offset sequenti provid greater coverag coupl natur interact augment realiti goal applic user inform base action physic world tour guid system good movement exhibit trigger displai inform applic typic albeit devic requir attent rest physic world metaphor describ interact user world similar tool electron geiger counter incorpor augment vision augment hear natur input voic close integr interact physic world resid starner macintyr system modifi user perceiv physic world figur lk figur system augment view requir headmount displai lightweight shown provid similar displai capabl tighter integr inform percept serendipit interact figur autom captur access live experi life busi academia spent listen event surround rememb import piec inform event clear potenti comput resourc augment ineffici human multipl stream relat inform virtual imposs captur manual tool support autom captur access live experi remov burden human good focu attent activ good indic good deal research relat gener captur access meet environ person earli work schmandt aron captur audio phone provid wai access content record treat audio uninterpret data success simpl techniqu provid inform overview live recent research effort captur type freeform ink tivoli suit softwar tool support meet moran electron whiteboard technolog liveboard support group discuss artifact produc electron whiteboard meet timestamp tempor inform meet index record audio provid richer set note meet similar integr record ink annot support classroom lectur greater emphasi autom captur materi access interfac larg popul student captur author fly lectur browser focu captur present attent captur arbitrari program interact video captur multipl sourc system focu captur group experi captur audio notebook al al focu captur individu system provid mix public person captur effort produc sort multimedia interfac review captur experi system provid autom support multipl camera integr present retriev mechan search larg repositori captur inform result access multimedia typic distribut web abowd review research commerci system emphasi ubiqu separ captur access phase electron captur move tradit brought closer user form interfac actual pen paper input form voic gestur accept treat raw data interpret provid understand captur experi challeng captur access substanti research advanc autom captur number open research summar separ issu primarili captur primarili access captur mention earlier import good drive applic ubicomp research captur main compel applic meet support compel applic area evid classroom point overwhelm accept captur student perspect explor captur equal compel record special event birthdai visit rel spend tim year reflect rememb event record film diari time camera preciou moment live record devic miss collabor design critic insight made inform set document properli technic exchang flow freeli opportunist encount formal design rich exchang inform discuss architectur poorli captur begun experi support captur inform brainstorm activ structur design meet mainten build support captur record actual construct contrast build plan repair technician construct mainten histori relev build artifact order determin repair except audio lectur work captur artifact physic world make easili access access phase emerg captur virtual lead research work area captur call raw stream inform captur direct playback analysi stream deriv addit inform simpl stream provid greater understand live event result analysi segment captur audio stream make predict topic commenc discuss chiu wilcox hierarch techniqu detect segment associ ink audio comput percept techniqu analyz simpl audio ink video signal applic signal analysi improv record raw stream autom product practic merg multipl camera feed coher high qualiti video view fix camera angl suffici captur salient part live scale system classroom entir afford pai technician sit classroom singl biggest challeng determin focu attent fo individu live event access access provid number playback capabl simplest playback situat inappropri overli ineffici review lectur student sit entir lectur pinpoint topic discuss replai portion summar experi glean salient point entir captur synchron multipl captur stream playback vital commerci stream emerg standard power synchron media stream clear product support stream user li ahead review stream user skim quickli point interest captur materi static reach access phase case annot captur materi note version problem comput numer challeng provid intuit interfac multipl version captur materi materi audio video timelin effect interfac manipul captur time captur artifact split number time timelin question newer interact comput good start point system move person system captur event public privaci concern captur access materi issu address specif design gener techniqu tag materi authent access discuss issu paper everydai comput describ emerg area interact everydai result consequ scale comput respect time push avail comput tradit desktop fundament relationship human provid continu interact move comput local tool constant presenc motiv everydai comput stem support inform unstructur activ typic everydai live activ continu constant flow action clear start end point familiar exampl orchestr commun famili manag inform design everydai comput requir address featur daili rare clear begin end fundament research activ point closur inform past recycl address book item basic activ commun inform manag design principl provid visibl current freedom dialogu simplic featur plai promin role interrupt expect activ albeit possibl oper conceptu resumpt activ start consist relat state prior interrupt interact model sequenc step resum built addit repres past interfac remind user action left multipl activ oper concurr activ multipl activ assum applic interfac user monitor background assist user resum activ resumpt base avail recent arriv inform design background interfac support multipl level convei monitor inform match rel urgenc import event room interfac present compel interfac spatial organ document applic multipl persist work space card standard met current commerci bar chang applic focu extens room provid awar assist user rememb past activ return room time import discrimin time fundament human measur stick rare repres comput interfac famili member week minut ago relev interpret incom call person search paper yesterdai month inform search numer wai incorpor time interfac edward regain work interfac repres past event length time interact applic interpret decid handl incom phone call react arriv local groceri util time inform tailor interact associ model inform hierarch model inform good match model inform activ princip associ inform multipl multipl perspect assum save email friend famili long time deal current attempt organ hierarchi folder topic organ result messag search vari success interfac list failur difficulti organ item list associ model organ support activ user inform numer point view view inher resum activ reason user retriev inform base current context enter offic arriv groceri store rememb inform rel current document edit week ago document colleagu circul similar topic comput ubiquit imper tool offer reflect role longer term activ principl everydai comput appli desktop design challeng relev continu chang user context mobil user shift activ comput resourc vari environ offic tool object plai multipl role activ comput augment whiteboard vari base contextu inform peopl present physic object paper file ambient displai provid entri point background inform activ distribut interact physic world implicit notion everydai relev research ubiquit comput synergi theme research everydai comput continu explor earlier interact focu design continu environ work tangibl media explor natur interfac support commun background awar current effort aim creat interact area support greater rang inform human activ respect audio aura relat previou tour guid chang locat trigger inform deliveri portabl devic motiv audio continu augment background auditori peripheri user ad dynam inform activ colleagu commun channel audio aura enhanc percept sphere inform user continu daili activ applic autom captur access move structur environ remembr agent retriev inform base physic context inform includ visual recognit user instruct system agent everydai inform continu index base physic locat bush design person memori record continu provid indic summari daili inform captur lam research direct everydai comput everydai comput offer challeng hci research commun current futur design continu present comput interfac multipl model portrai comput ubiquit model wholli satisfi notion inform applianc typic reflect devic sit background knowledg activ interfac borrow tradit gui concept consum electron comput system continu oper learn past activ act typic repres anthropomorph agent doubt interfac base dialogu talk head person research wearabl explor limit current input displai technolog typic rudimentari interfac present inform level peripheri human attent interest tangibl media collabor work wearabl current interfac typic present gener backdrop mechan background move inform foreground attent current design experi aim creat interfac oper level peripheri connect event physic virtual world peopl oper disconnect virtual space document web page physic space book paper file human activ coordin space effort earli digit desk work left understand combin inform space match peopl conceptu endeavor modifi tradit hci method support design opportunist behavior methodolog understand role comput everydai live combin inform method laboratori experi ethnograph observ simpl research classroom goal learn interrog result deriv evalu strategi chosen spectrum method match question learn method inform result combin effort work continu discuss section evalu ubicomp system addit challeng ubicomp import topic ubicomp research evalu social implic cut theme address evalu ubicomp system order understand impact ubiquit comput everydai navig delic balanc predict technolog serv real human observ authent subsequ human activ technolog form summ evalu ubicomp system difficult discuss challeng publish evalu perspect ubicomp commun notabl except work publish xerox parc research tivoli captur system context technic meet al research ubiquit comput limit impact hci commun respect advic undertak challeng find human major difficulti evalu ubicomp system simpli reliabl system evalu technolog creat ubicomp system cut edg understood difficult creat reliabl robust system support activ continu basi good portion report ubicomp work remain level demonstr prototyp kind research critic good feasibl research technolog import ubicomp research research build compel system infrastructur built technolog serv real perceiv human weiser ubicomp provid applic serv human compel stori simpli provid demonstr vehicl research result provid basi evalu impact system everydai life intend popul situat build compel stori activ continu basi creat live laboratori work continu motiv constant feedback lead understand design system mistak made difficult challeng build robust ubicomp pai price build sophist infrastructur find fall short address goal set compel stori sort feasibl studi applic sink substanti effort engin robust system deeper evalu feasibl evalu driven perspect goal determin system kind activ user engag reaction posit neg answer question inform futur design futur evalu plan import understand system intend popul perform quantit studi impact case xerox design ubiquit comput applic requir design project futur user emploi technolog design imposs interact hci issu exacerb impli paradigm shift hci result distribut comput capabl physic environ design work al emploi ethnograph observ whiteboard coupl questionnair understand peopl whiteboard daili basi figur rich data observ inspir design work constraint notion portion board user expect chang result observ real whiteboard data observ kei ground user studi questionnair interview discuss easili slip user refer week observ uncov examin detail daili practic technolog augment whiteboard readi user gather wealth inform observ interview critic inform design case audio afford usabl issu input output technolog understood introduc technolog set evalu influenc subsequ design design audio aura al interest explor awar relev offic activ enhanc ambient sound mobil combin activ wireless headphon audio gener real adopt user headphon social prohibit cover ear black shell figur lk figur design observ whiteboard inform design whiteboard draw board basi discuss complex concept illustr detail bottom slide capabl develop control sound present limit creat rich auditori space understand potenti interact technolog limit remov futur emploi scenario base inform observ xerox parc work guid design evalu scenario incorpor inform peopl parc work includ practic gather drop offic physic build long hallwai backbon layout ground scenario common potenti user reflect daili activ evalu design scenario understand interact time commun path compon technolog fast meet interact demand speed modif requir balanc set speed import design goal scenario understand design space explor evalu context authent deeper evalu result requir real requir authent set scale dimens ubicomp system peopl tim make imposs usabl laboratori effect user observ interact system routin requir realist environ expect case classroom case center georgia demonstr research project evolv earli prototyp feasibl studi matur system larg popul live classroom shown figur project began juli intent produc system captur classroom experi facilit review student figur lk figur classroom abil learn actual georgia tech educ environ teacher student head write hear futur refer write activ process cue felt desir student teacher perspect afford opportun student lift head engag lectur experi captur system reliev burden test feasibl hypothesi month project provid environ captur entir observ initi hypothesi worth test learn valuabl lesson extend experi initi experi includ student devic clear distract student abandon part resum past month technolog caught learn initi experi order understand impact captur system teach learn gather usag data larger set class requir engin effort creat robust reliabl captur system spring quarter support multipl class simultan captur gain insight system futur direct direct result deeper system encourag user modifi behavior modifi behavior good learn practic reinforc time reintroduc student unit person captur experi encourag practic facilit retriev synchron playback lectur experi insight motiv research effort establish research stand model ubicomp research autom captur access evalu techniqu inappropri assum authent user comfort servic provid develop habit question appli qualit quantit evalu method major usabl techniqu task evalu perform determin fit system interfac complet task clear appli evalu techniqu inform everydai comput situat social issu ubiquit comput push make easier comput understand react phenomenon physic world record phenomena enabl technolog carri numer make easi peopl build system effect spy control author ubicomp research work understand issu fear call work work design social solut address concern basic concern inform store comput access modifi content secur encrypt scheme part technolog solut inform glean environ transport network work wearabl comput design approach provid secur bit local remov risk transport public network fear user lack knowledg comput system origin vision ubiquit comput comput disappear physic counter inform user design solut emploi make inform visibl system physic phenomena captur live situat provid clear indic record occur record capabl commonli challeng everydai comput enabl peopl awar peopl ascertain visibl physic space public cue convei visibl virtual space step record control stop activ control distribut inform challeng relat design collabor environ action role collabor fluid difficult articul static snapshot inform determin time specif practic peopl workplac home set number reaction system builder handl sensit topic captur xerox solut captur captur summari portion technic meet classroom default record attempt obtain good qualiti audio video lectur front room system control portion audio record futur refer solut present mean reduc amount storag requir merit approach perspect enabl perfect captur system part past interest challeng collabor situat figur accept polici share memori posit slant issu focu wai accommod awar valuabl segment captur experi lieu elimin forget part captur histori issu surround dissemin inform dawn human specif concern stem ubicomp make kind inform gener fact comput easili track daili feat requir larg amount human addit address mention concern visibl approach creat design exampl benefici inform afford low qualiti video earli media space amount inform convei social imag met import awar feel connect violat privaci concern design audio aura great care convei qualit inform activ colleagu stop inform obtain system person offic akin inform neighbor offic provid social issu directli link privaci record meet lectur posit neg impact attend posit knowledg record encourag peopl commentari neg knowledg peopl contribut discuss fear futur subtl problem notic extens experi classroom student chose question class answer discuss student back listen lectur social legal practic continu evolv concert technolog design innov situat peopl compar perceiv benefit cost ubicomp technolog wear radio locat rescu personnel benefit understand locat recent research detail calendar practic sun microsystem colleagu share extens inform daili collabor activ discuss previou section understand social implic technolog peopl invent technolog sand shift beneath attent issu control privaci ensur posit technolog conclud thought attempt outlin trajectori ubicomp research decad inspir work weiser colleagu xerox parc identifi research theme provid background achiev area highlight remain challeng desir motiv ubicomp research attack import problem doubt left import challeng ubicomp forward problem articul solv weiser claim point ubiquit comput creat compel applic drive develop devic infrastructur spirit promot broader view promot gener util ubiquit interact comput resourc applic focu fruit hci research success comput case singl applic driven critic mass accept motiv applic person comput offic palm person real goal ubicomp provid interact promot unifi continu interact human comput servic focu human time singl interfac accomplish task interact akin interact rich physic world place object everydai live acknowledg author acknowledg inspir late mark di suddenli initi draft paper vision enthusiasm influenc countless number indebt contribut wai acknowledg interact influenc number colleagu xerox parc georgia futur comput environ group colleg comput refer cg pinkerton mobil tour guid acm wireless vol pp investig integr access problem ubiquit comput educ set chi lo pp autom visual multipl media stream ieee confer multimedia comput system san pp gd classroom experi instrument live educ environ ibm system special issu focu comput chr tool servic author fly confer educ multimedia june audio augment prototyp autom tour guid acm short kn support captur access interfac inform opportunist meet gvu technic januari atlant volum rv notic differ speech recognit acm read inform vision morgan kaufmann jm mari beth view matcher design space cheverst mitchel davi design object model context sensit tourist guid interact applic mobil comput novemb wilcox dynam group techniqu ink audio acm symposium user interfac softwar technolog schilit lightweight note share group pp wk ed techniqu autonom collabor acm pp larg interact displai support group present remot collabor chi pp pp augment realiti commun altern desktop metaphor acm pp lai foundat graspabl user interfac acm pp sh virtual realiti palmtop comput special issu virtual world acm transact inform system declin accuraci automat speech recognit function time fatigu voic intern journal machin recognit accuraci user accept pen interfac acm emerg groupwar success major studi adopt adapt squeez hold tilt explor manipul user interfac sk multipl virtual workspac reduc space content graphic user interfac acm transact vol ubiquit captur spontan collabor nov tangibl seamless interfac bit atom augment experi probabilist vision wearabl ieee intern symposium wearabl oct intim comput support human memori technic report rank intern symposium gener human interfac ministri intern trade industri futur multimedia user interfac multimedia ed intellig augment realiti interfac intellig environ aaai spring symposium intellig environ march stanford ca provid integr support ambigu interfac van tool captur access collabor activ multimedia pp van find implicit structur spatial layout inform pp van evolutionari engag ongo collabor work case studi pp van interact techniqu organ materi electron whiteboard acm symposium user interfac softwar technolog pp van case studi multimedia meet record pp passiv captur structur lectur acm confer multimedia pp design audio aura acm confer human factor comput system lo dimens offic whiteboard acm confer human factor comput system ed write wall interact chri johnson publish io ifip da good product person comput inform applianc solut mit press adapt support mobil museum guid interact applic mobil comput novemb perceptu user interfac workshop octob perceptu user interfac workshop san nov affect comput mit press approach inform environ acm symposium user interfac softwar technolog pp world comput augment interact real world environ virtual augment realiti acm symposium user interfac softwar technolog pp bj wearabl remembr system augment memori person technolog journal special issu wearabl person technolog bj te remembr continu run autom inform retriev system practic applic intellig agent technolog london gd autom captur retriev architectur technic note work ifip confer softwar san februari full paper georgia tech gvu center technic report ak gd context aid develop applic acm pp graphic telecommun interfac societi inform rb aspect recogn iee colloquium handwrit number institut electr london augment realiti wearabl presenc lj augment audio notebook companion short paper pp lj audio paper pen interact structur speech phd mit media august interact landscap creativ innov pp ja person captur public experi acm symposium user interfac softwar technolog pp activ badg locat system acm transact inform parctab ubiquit comput experi xerox palo alto research technic report march tool video log pp comput centuri scientif sept comput scienc issu ubiquit comput commun juli interact paper digit desk commun acm handwritten note provid access record dynam organ ink audio notebook investig error recoveri dialogu user interfac manag system speech ifip commun inform retriev meet support tool pp 
HCI/144 4 short titl fundamenta informatica io press qualit spatial reason overview cohn school comput leed unit kingdom abstract paper overview major qualit spatial reason techniqu survei main aspect qualit knowledg includ ontolog orient shape qualit spatial reason includ reason spatial chang final discuss theoret result futur work paper version qualit spatial ontolog text slightli di erent format fi format cohn overview introduct qualit reason concern captur everydai knowledg physic myriad equat engin scientist explain complex physic phenomenon creat quantit model princip goal qualit reason make explicit reason comput make explain behaviour physic system qualit manner intract unavail quantit model essenc qualit reason wai repres continu properti world discret system symbol quantiz quantiz equal state idea relev distinct made quantiz relev kind reason perform result set qualit valu term quantiti indistinguish valu identi ed class natur order partial quantiti form simpl ectiv infer exploit transit order qualit arithmet algebra produc ambigu answer research qualit reason literatur devot overcom detriment ect search space result ambigu import aspect work tradit qualit reason worth note standard assumpt chang continu simpl consequ quantiti pass intermedi valu frequent quantiti space variabl transit rst notion exploit work qualit spatial reason qualit spatial reason spatial dai interact physic case driven qualit abstract complet priori quantit knowledg qr hold develop theori reason space justi es interest studi spatial concept cognit point view provok birth note qualit reason infer made absenc complet probabilist fuzzi techniqu reli arbitrarili assign probabl membership di erenti quantiti sucient evid case indistinguish relat relat import characterist perceptu seri small combin form percept chang indistinguish relat exiv lead paradox percept place theori indistinguish theori percept present wai avoid paradox word theori context set formal axiom properti relat collect natur scientist empir testabl explan observ regular cohn overview qualit spatial reason ai gi space adequ repres singl scalar quantiti natur space qr earli naiv physic manifesto neglect partial attribut conjectur pure qualit success reason physic system success develop pure qualit reason spatial kinemat mechan conjectur fact imposs pure qualit spatial reason mechan forbu al correctli identifi transit valu kei featur qualit quantiti space doubt exploit higher dimens conclud space higher dimens spatial reason weaker number challeng provid calculi machin repres reason spatial entiti resort tradit quantit techniqu preval comput graphic comput vision commun amount research coupl year weaken conjectur qualit spatial address di erent aspect space includ size distanc put forward rich exploit transit demonstr composit tabl tabl allen interv tempor logic built paper overview major qualit spatial reason techniqu section mention applic qualit spatial reason section survei main aspect qualit spatial knowledg includ ontolog orient shape section discuss qualit spatial reason includ reason spatial chang paper conclud present theoret result futur work paper version applic qualit spatial reason research motiv wide varieti applic area includ geograph inform system robot high level spatial proposit semant natur engin reason physic system visual languag syntax semant numer applic area includ qualit recognit notion biologi domain space gi major problem interact gigabyt inform store vector raster cientli support intuit orient interact user abstract mass numer data queri qualit gener gi built concept naiv geographi naiv geographi emploi qualit cohn overview reason combin provis queri robot navig ultim requir numer speci ed direct robot move hierarch plan detail delai high level plan achiev shown ectiv robot model environ lead inabl standard robot navig techniqu qualit model space facilit plan approach develop robust qualit method robot map navig spatial environ qualit solut mover problem comput vision visual object recognit higher level includ interpret integr visual inform qualit symbol reason shape object match geograph object techniqu interpret result comput higher level descript scene video input qualit predic ensur scene semant close ident similar descript natur interpret spatial proposit tend ambigu multipl wai natur languag spatial preposit cite di erent mean li motiv qualit spatial nding formal describ preposit engin robot ultim requir fulli metric descript earli stage design reason qualit descript suce eld qualit kinemat larg concern support type activ visual visual program languag kind lack formal speci cation kind expect textual program languag languag make metric bulk predominantli qualit exact length compon diagram pictur unimport import topolog relationship compon li theori qualit spatial reason languag aspect qualit spatial repres space rich histori physic scienc serv locat object quantit framework spatial express tend oper partit domain descript space ad hoc basi emerg qualit spatial reason partit systemat di erent aspect space decid kind spatial entiti admit commit ontolog develop di erent kind wai describ relationship kind spatial size distanc rel orient shape cohn overview ontolog mathemat theori point consid primari primit spatial entiti point extend spatial entiti region de set point minor tradit philosoph error strong tendenc region space primit spatial entiti ontolog shift mean build theori spatial geometr strong reason region ontolog primit interest spatial theori reason physic argu spatial extens physic object lower dimens entiti de ne term region admit time advantag view physic entiti entiti entiti dimens pertin question mix dimens entiti discuss issu found ontolog question natur spatial imagin applic discret nite convex tension model favour approach handl spatial inform digit lower level attempt bridg gap develop qualit spatial theori base discret model space recent investig discret continu decid ontolog primit logic amount decid primit logic symbol admit de constrain set axiom argu set primit mathemat eleg make easier assess consist simplifi interfac symbol system compon fewer primit implement argument result symbol infer complic natur larg rich set concept mean axiom connect di erent wai nal ontolog question model multi approach model space dimens project region dimens reason dimens separ approach object overlap project ax fact overlap domain orthogon align object simon perceiv peopl perceiv individu nite cohn overview topolog topolog fundament aspect space clear topolog form fundament aspect qualit spatial reason topolog make qualit distinct topolog studi extens mathemat abstract relev attempt spatial reason qualit spatial theori uenc mathemat number reason import undesir gener interest qualit spatial reason paid attent mathemat algebra topolog reason form region relat region region relat region relat hold adapt convent mathemat strategi adopt exist approach topolog commun work found philosoph logic commun work built axiomat theori space predominantli topolog region point primit tradit term work clark lead develop call system develop ec dc po figur illustr relat calculu continu transit neighbourhood clark primit notion idea region connect region consist set system interpret slightli chang closur region share point theori present suitabl distanc mean distanc cf ect distinct closur argu relev kind domain concern surprisingli de ne predic function captur interest topolog distinct set jointli exhaust disjoint relat illustr figur set formal semant canon cohn overview model arbitrari ground boolean atom procedur gener actual interpret ensur individu connect principl supplement hold work mention base clark calculu origin interpret fusion oper made rst order error correct contrari argu di erenti closur interior cognit materi object close individu complement open interpret share point framework di erent notion connect literatur analyz present dimens variat closur region connect point connect compon region express toru solid hole doubl loop block minu block solid gap figur distinguish shape taxonomi topolog properti relat de ned singl predic simpl predic de ne predic de ne predic count number time region touch seri set task distinguish solid piec region singl shown restrict set assumpt topolog properti region target region shape depict figur distinguish rang topolog distinct region made shown figur interior connect mean interior region piec easi express properti term rule region exclud region figur boundari simpl curv form point neighbourhood boundari local de ne predic cohn overview rule case figur mean straightforward demonstr de nition intend con figur di erent type con region sourc diculti fact region model axiom assum uniform fact region refer directli boundari region share boundari ec relat entiti di erent tendenc work involv qualit spatial reason spatial entiti consid theori dimens case reason relat system base predic power inch calculu treat point spatial extend entiti special gener notion extent aim improv express calculi avoid counterintuit consequ approach argu standard model space time base segment wrong topolog structur event region boundari illustr modifi model topolog structur address problem repres reason region di er mix variant classic origin work part relat topolog boundari boundari region dimens lead desir intuit pictur strictli linear hierarchi dimens theori introduc notion boundari region explicitli includ topolog altern approach repres reason topolog relat seri paper set point region boundari complement relationship region matrix call task trivial primit introduc section reason region di erent imposs sort structur essenti copi theori simpler matrix featur interior boundari sucient describ relat matrix express set relat de ned note account relationship region space cohn overview account physic realiti space speci assumpt natur turn remain matric relat calculu reason region hole classifi relationship pair relationship hole region region hole di erent calculi relat deriv chang underli assumpt region matrix repres intersect deriv calculu repres reason region extend matrix cell dimens intersect simpli exist relat line point larg number relationship de ned term base gener set ve polymorph binari relat pair spatial entiti cross complex relat form conjoin atom proposit form ve relat argument boundari endpoint oper appli express calculu cbm combin relat relat give total relat mode overlap varieti wai region partial overlap previou work except partial overlap singl relat denot connect singl relat whilst realiz potenti nite varieti partial overlap matrix number connect compon investig matric number greater theoret physic topolog present systemat account subtl relat topolog note whilst sucient theori literatur integr topolog main strategi integr cohn overview gener ad topolog primit al add topolog primit connect spatial entiti part relat altern singl primit connect part main advantag separ theori topolog colloc share part approach topolog primal theori topolog theori base mention de ne eleg singl uni ed coloc impli share part theori lower spatial absolut nal approach topolog introduc domain speci theori addit primit introduc idea restrict quanti cation introduc predic de ned def topolog fulli metric spatial topolog abstract qualit spatial potenti situat inform insuci section explor di erent wai qualit inform repres orient orient relat describ object rel de ned term basic primari refer object frame refer unlik topolog relat spatial entiti orient binari relat orient primari respect refer kind frame manifest displai qualit orient calculi found calculi explicit relat frame refer explicit relat worth mention work earlier work develop calculu base function map tripl point qualit denot collinear orient reason visibl locat qualit navig task shape descript develop calculu reason rel orient pair line segment orient base relat true show number qualit calculi translat system reason system constraint logic exploit disadvantag theori reason np complet nement lead algebra ternari relat cohn overview cyclic order orient atom gener relat algorithm polynomi complet subclass includ atom relat distanc size spatial distanc divid main measur provid kind rel measur tradit qualit reason primarili concern deal linear quantiti qualit algebra transit quantiti space mention earlier distanc size measur interest context order magnitud calculi develop qr commun tradit qr formal kind delta calculu introduc note amount term big speci calli develop qualit spatial reason earliest primit true bodi connect simpl translat rotat shape primit easi de ne notion nearer farther primit simpl size metric region de region larger connect region method determin rel size object reli translat region shape size exploit topolog relationship translat region proper part smaller distanc close relat notion distanc sum distanc point region vari depend orient number calculi base primit combin distanc orient inform straightforward idea combin direct repres segment compass simpl distanc metric slightli sophist idea introduc primit de ne posit point respect direct line segment point approach combin knowledg distanc posit qualit combin orient present liu explicitli de ne semant qualit distanc qualit orient angl formul qualit de ne composit tabl calculu combin type inform suggest comput quantit visual simul anneal interest framework repres distanc complex result calculi present section straightforward rel measur simpli write cohn overview tend includ orient distanc system order sequenc distanc relat set structur relat give addit inform distanc relat relat distanc accept distanc success accept area de ne sequenc structur relat de ne relationship typic structur relat monoton properti greater sum structur relationship order magnitud structur relationship import ning composit tabl homogen distanc system distanc relat structur case heterogen distanc system system fact context ect distanc handl di erent frame distanc system infer di erent frame refer articul rule obviou ect move context qualit distanc term vari distanc behav wai context distanc symmetr domain shortest distanc point straight line lake build shape shape import characterist dicult describ qualit pure topolog theori limit statement made shape interior piec observ constrain shape rigid object topolog constraint relat applic demand ner grain kind inform introduc explicit qualit shape descript introduc kind shape primit whilst retain qualit mathemat commun develop di erent geometri express euclidean project necessarili develop ecient comput reason techniqu qualit approach shape approach work describ boundari object includ classifi sequenc di erent type combin qualit orient absolut distanc combin qualit orient rel distanc inform combin distanc posit calculu section introduc composit tabl distanc measur time journei longer return downhil journei orient distanc primit discuss add pure mention larg point base directli applic describ shape region cohn overview ari segment describ sequenc di erent type curvatur extrema contour techniqu slope project approach describ polygon corner describ qualit direct corner set approach altern construct complex shape region simpler line construct solid start qualit set primit dichotomi drawn primarili describ boundari object compar repres interior arguabl techniqu prefer shape inher concept version medial axi de ning grain region lead modi ed version tradit call angular skeleton shape descript volum compon exempli ed work shape abstract primit bound box convex hull consid brie model whilst techniqu investig extens calculu theori shown interest predic de ned notion convex hull region predic test combin topolog comput topolog relationship shape di erent compon di erenc convex hull distinguish di erent kind concav shape nement techniqu exploit idea shape descript describ convex compon di erenc convex hull shape convex hull power primit fact recent shown system essenti compact planar shape relat transform distinguish constraint languag leav topic shape descript point work describ shape properti compact elong minimum bound rectangl shape order magnitud calculu elong comput ratio side minimum bound rectangl whilst compact compar area shape minimum bound rectangl notion voronoi hull uncertainti granular uncertainti granular featur inform domain knowledg uncertainti vagu indetermin region boundari granular notion gener descript requir case vagu spatial concept distinguish spatial situat object region occupi adequ treatment vagu spatial theori abstract detail gener inform receiv attent ai commun cohn overview inform account vagu region vagu relationship philosoph debat vagu object exist formal theori deal vagu extent glue standard numer techniqu reason uncertainti research extend exist qualit spatial reason techniqu explicitli repres reason uncertain inform extens calculu similar approach approach continu seri paper postul exist crisp region addit crisp region add binari relat region varieti relat de ned term primit extend theori relat origin repres reason ind nite region boundari model region indetermin boundari pair region shown figur repres minimum extens ind nite whilst egg repres maximum extens turn gener relat crisp region natur cluster set egg crisp figur structur region indetermin boundari extens similar calculu ne granular postul cluster group ad hoc techniqu achiev interest extens work show calculu region broad boundari reason region indetermin boundari cover number kind region includ convex hull minimum bound bu er zone raster notion ind nite relat locat deal notion part rough locat spatial object exact locat region space object notion part locat introduc relat part spatial object part spatial region rough locat spatial object part locat spatial object respect set region space form region partit notion rough locat link part spatial object part partit region argu observ measur locat physic realiti yield knowledg rough vagu de ned object locat region special applic model raster undertaken cohn overview partit consist concentr boundari notion rough locat partit consist concentr region coincid notion vagu region introduc calculu repres reason locat rigid object move region present present de ne notion mobil describ subset relat suitabl repres spatial analyz comput complex set work base congruenc worth note similar idea rough set exact relationship explor approach spatial uncertainti work indistinguish relat transit fail gener class develop nonmonoton spatial logic qualit spatial reason work concentr aspect comput paradigm investig includ constraint base reason preval form qualit reason base composit tabl composit infer relat fact form relat fact form involv valid composit infer depend case constant involv logic properti relat case composit pair relat maintain tabl requir techniqu signi canc deal relat inform involv xed set relat set store composit tabl relationship pair relat entri disjunct qualit natur calculu limit set binari simplic composit infer make attract mean ectiv reason composit tabl receiv attent calculi mention paper composit tabl construct time pose challeng reformul rst order logic order intuitionist proposit formula repres constraint situat gener composit tabl shown tabl automat composit tabl necessarili subsum form desir reason interest question composit tabl reason sucient infer origin allen analysi tempor relat call tabl appropri renam tabl relat involv composit relat repres transit individu relat reformul interest true spatial theori spatial implic interpret disjunct sum region approach automat gener composit tabl order modal logic cohn overview ec dc dc ec po dc eq dr dc dr dc dc po eq po dc ec pp dc ec po eq dc dc tabl composit tabl relat mechan theori suggest formal relat algebra suit repres composit reason deserv studi wider applic case composit base reason gener constraint base reason sucient constraint satisfact problem resort theorem kind theorem prove reason spatial chang concern pure static spatial calculi import develop calculi combin space time integr fashion topolog spatial entiti chang dimens abstract granular shift spatial chang chang number topolog compon break fuse blob chang number tunnel drill block chang number interior put lid domain assum chang continu case tradit qualit requir build spatial calculu respect underli continu natur requir common di erent kind spatial chang mention import qualit valu relat neighbour predic hold continu chang predic hold neighbour continu network de ning neighbour call conceptu neighbourhood literatur consequ treatment aspect space orient zimmermann approach cardin direct frank approach qualit distanc integr spatial model view road line entiti wide entiti rang mountain pass delic cargo chang polit move boundari entiti discontinu manner cohn overview term describ structur allen relat conceptu close meet neighbour overlap illustr figur qualit spatial calculi report paper conceptu neighbourhood construct qualit simul common form comput tradit qr literatur qualit conceptu neighbourhood diagram easi build qualit spatial simul simul set ground atom statement describ initi state construct tree futur state branch tree result ambigu qualit calculu continu provid sucient constraint restrict gener state reason set gener domain speci constraint requir addit desir extens analog earlier qr work incorpor proper theori spatial process couch languag work direct eld base theori spatial process heat ow analysi topolog conceptu neighbourhood diagram correspond process rotat region process resist boundari base shape descript languag shape infer sequenc process chang worthi note qualit spatial simul work base system qualit motion larg amount work theori basi work motion qualit framework motion framework work insist concept dynam research default reason metric space motion kei notion understand spatial relat relat form potenti continu remain implicitli assum notion problem conceptu neighbourhood built manual calculu potenti error prone oper relat techniqu arriv automat analysi structur conceptu neighbourhood report goal fundament approach exploit continu underli semant space note lift notion close individu relat entir scene set relat common object gain measur conceptu similar suggest close relat notion topolog predic neighbour respect matric di er fewer entri result neighbourhood graph ident true conceptu neighbourhood continu graph link miss construct simul transit calculu approach cohn overview construct conceptu neighbourhood class relat infer relat domin domin hold interv instantan domin eq domin neighbour relat domin analog equal chang law found tradit qr stricter tempor order event occur qualit simul motiv desir exploit decid modal logic qualit seri express calculi easili repres restrict continu motion approach automat infer continu network muller idea histori enrich theori intend spatial entiti achiev formal theori reason motion present model primit entiti tempor relat de ned express power theori de nition complex motion class express motion verb natur languag properti assum desir part theori motion recov give sound theoret basi qualit motion altern approach explor express power qualit geometri problem repres reason motion rigid bodi con ning environ theoret result work gener spatial remain number theoret question calculi formal semant inventor question simplest semant semant task show set formula consist set entail complex procedur theori weak true formula stronger formula made true theori theori decid properti model theori model model di erent cardin de nition interest properti categor countabl model model user viewpoint question special region base theori ontolog answer standard point base interpret approach base approach signi advantag firstli assum morpholog primit sphere predic congruenc emploi addit topolog primit topolog concept de nabl prove categor direct encod tarski geometri axiom languag cohn overview simplest model altern model involv region nite piec argu region base theori interpret point set theoret fundament result decid widespread applic qualit spatial theori show boolean algebra ad closur oper extern connect relat result undecid system encod arbitrari statement arithmet impli clark calculu relat calculi rst order theori calculi undecid question decid subsystem constraint languag shown decid achiev encod relat set formula intuitionist proport calculu decid calculu languag subsequ shown tractabl fact satisfact problem solvabl time complex class nc constraint languag constraint arbitrari disjunct tractabl maxim tractabl subset constraint languag identi ed shown path consist sucient decid consist case size constraint introduc region reason ectiv polynomi complet classi cation tractabl made turn maxim tractabl subset relat subset set base relat work complex includ complex topolog theori present orient calculu determin polynomi subset base whilst determin sati abil gener algebra np complet similarli show whilst gener consist problem algebra cardin direct np consist relat polynomi set maxim tractabl subset complex analysi morpholog calculu base congruenc interest analysi consid calculu simpler calculi determin number di erent problem instanc relat consist planar realiz tractabl harder problem shown constraint languag intract hard determin set algebra constraint real clark system semant set euclidean space shown complet weak turn contrari clark relat theori fact complet boolean algebra system correct problem clark mention semant shown complet author manner descript logic commun sought line divid decid undecid tractabl intract cohn overview notion connect forc standard model model complet result strong deriv complet topolog theori element nite region primit null boolean function predic test region piec theori rst order requir rule infer view undecid rst order topolog mention rule infer exist model region sum nite connect region result theori complet decid notwithstand attempt deriv complet rst order topolog clear rst order nite topolog complet categor decid final comment issu address literatur issu cognit valid claim made qualit reason akin human empir justi cation except studi made calculu repres topolog relat region line studi investig prefer allen relat spatial case composit tabl entri disjunct fact human prefer model explain reason ecient presenc kind ambigu qualit extend evalu egenhof approach topolog relat paper survei kei idea result space exhaust survei eld knowledg singl spatial languag emerg hope eld develop librari reason devic criteria success applic case spatial qualit quantit knowledg reason qualit quantit reason complementari techniqu research ensur integr develop reliabl wai translat kind formal interfac symbol techniqu develop diagrammat reason commun interest import challeng enforc model lead question good idea model distinct strong weak connect topolog pure spatial appli theori physic object materi substanc region occupi extens theori spatial granular ner ner granular fewer instanc weak connect true limit theori classic topolog model exist research problem includ cohn overview acknowledg rst author gratefulli acknowledg nancial support epsrc grant recent eu support network workshop provid excel forum discuss idea paper contribut commun colleagu gratefulli acknowledg author acknowledg nancial assist refer receiv scholarship unit kingdom refer knowledg tempor commun page nonmonoton nebel advanc arti cial page geometri common semant complet axiomat proc intl joint conf arti cial intellig page approach semant semant preposit mental process natur languag de del de edit del logic incid journal logic page reason proposit torasso principl knowledg reason proc intl conf san morgan step construct absolut complet theori spatial jj lm pereira orlowska proc page logic qualit spatial bulletin interest group pure appli logic page ftp mpg index html composit tabl provid complet tractabl proof procedur relat constraint proc workshop spatial tempor foundat qualit proc european conf ai cohn overview rigid bodi motion qualit theori spatial proc american conf ai notr dame journal formal page theori human imag psycholog page ontolog epistemolog rough mark spatial inform theori cognit comput foundat geograph inform proc lnc page pointless theori space base strong connect aiello doyl principl knowledg proc intl conf morgan page human machin spatial proc intl symposium spatial data franci page structur spatial philosoph page hole super mit de lenzerini express descript robinson voronkov handbook autom elsevi scienc publish descript recognit multiresolut imag vision page calculu individu base notr dame journal formal page notr dame journal formal page di small set formal topolog relationship suitabl end user abel ooi advanc spatial proc intl symposium spatial lnc page topolog spatial strategi queri comput page di comparison method repres topolog inform page di algebra model spatial object undetermin frank proc specialist meet geograph entiti undetermin franci di topolog intern journal approxim cohn overview di global framework qualit shape di posit arti cial page hierarch qualit space base connect frank kuhn spatial inform theoret basi proc lnc page spatial reason nebel advanc arti cial lnai page spatial proc workshop adapt spatial dynam page reason qualit spatial relat stock spatial tempor page comparison structur spatial tempor smith white philosophi cognit scienc proc intl wittgenstein vienna region undetermin proc workshop theori spatial region indetermin smith topolog foundat cognit region indetermin frank proc specialist meet geograph entiti undetermin franci page approach repres spatial aiello shapiro principl knowledg proc intl conf morgan page calculu region base qualit spatial page logic de ned qualit spatial intern journal special issu formal ontolog conceptu analysi knowledg page relat prade proc european conf ai wilei page taxonomi qualit topolog mark spatial inform cognit comput foundat geograph inform proc lnc page complex reason spatial journal cial intellig page cohn overview locat cohn giunchiglia selman principl knowledg proc intl conf morgan page simul base logic formal space proc american conf ai page paradox percept limit levesqu reiter principl knowledg proc intl conf morgan page morgan network topolog relat page de line surfac set journal page de measur rough set arti cial page approach region connect technic school inform softwar voronoi model cultur applic social scienc frank spatial inform theoret basi proc lnc page shape tangenti axi transform shape web www paper grain html formal de nition binari topolog litwin schek intl conf foundat data organ lnc page proc workshop foundat cognit volum report doctor seri cognit process journal visual languag page topolog spatial intern journal geograph inform page topolog spatial relationship line area formal natur languag spatial predic technic report nation centr geograph inform santa gradual topolog theori method reason geograph lnc page cohn overview di relat region intern journal geograph inform page topolog intern journal geograph inform page frank kuhn spatial inform theoret basi proc lnc page relationship region abel ooi int symposium larg spatial lnc restrict intern journal page framework base clp extend chr reason qualit orient posit journal visual languag page qualit spatial reason theori practic applic robot navig frontier ai volum io vagu reprint collect clarendon symbol approach qualit arti cial spatial reason algebra frank kuhn spatial inform theoret basi proc lnc page ag qualit event model automat video imag vision page topolog arti cial page reason space gentner steven mental page process arti cial page kinemat proc intl joint conf ai page spatial reason distanc direct geograph journal visual languag page reason base arti cial page util spatial structur cognit plausibl ecient proc ieee intl conf man page reason document technic texa cohn overview oper qualit spatial frank kuhn spatial inform theoret basi proc lnc page integr logic time proc intl joint conf ai qualit theori frank kuhn spatial inform theoret basi proc lnc page dimens qualit spatial wahlster proc european conf ai john wilei page time stock spatial tempor page journal visual languag page discret mark spatial inform theori cognit comput foundat geograph inform proc lnc page topolog qualit size constraint spatial proc intl conf principl practic constraint program handbook incid elsevi page theori arti cial page diagrammat mit spatial logic describ visual arti cial intellig paper integr natur languag vision process qualit proc intl workshop qualit aaai page languag syntax spatial logic marriott meyer proc workshop theori visual sequenc order type discret comput trend discret comput page ning made smith topolog foundat cognit volum report doctor programm cognit scienc de ning connect torasso principl knowledg proc intl conf morgan cohn overview axiomat approach topolog spatial inform technic report school comput inch proc intl symposium arti cial intellig singl primit de ning topolog properti relat term technic report school comput formal describ topolog spheric technic report school comput connect base approach topolog descript page proc intl joint conf ai morgan page topolog fundamenta page constraint base approach reason journal appli page fulli formal theori describ visual marriott meyer visual languag pp reason descript schubert shapiro principl knowledg proc intl conf morgan page uniform time space mutual constraint comput math ve physic expert system microelectron edinburgh ve physic ontolog hobb moor formal theori ablex publish page de ning motion verb spatial universitat de spatial lectur note arti cial intellig springer di frank kuhn spatial inform theoret basi proc lnc page languag spatial cognit interdisciplinari studi preposit cambridg proc intl joint conf ai page algebra cyclic order proc american conf ai page cohn overview spatial algebra model rel fachbereich complet classi cation tractabl journal arti cial intellig page www cs washington ps spatial reason object shape qualit frank spatial inform theoret basi proc lnc page formal theori spatial principl knowledg proc intl conf page cognit assess topolog spatial result empir spatial inform theoret basi proc lnc page mental model qualit spatial cognit assess allen proc annual conf cognit scienc lawrenc erlbaum page robot explor map strategi base semant spatial journal robot autonom page reliabl semant data integr sort proc conf inform knowledg acm page process grammar arti cial page gener conceptu neighbourhood tempor spatial anger proc workshop spatial tempor cardin journal visual languag page method spatial reason base qualit arti cial page qualit model gradient ow spatial distribut proc intl workshop reason page evalu ning comput model spatial relat spatial inform theoret basi proc lnc page recognit spatial organ proc royal societi seri page nite mark spatial inform theori cognit comput foundat geograph inform proc lnc page cohn overview reason process comput chemic page survei comput model spatial express oliv process spatial qualit model proc american conf ai vol page dune du pour la du de sen phd universit paul qualit theori motion base schubert shapiro principl knowledg proc intl conf morgan page primit space guarino formal ontolog inform proc int conf frontier ai volum io page properti qualit spatial advanc arti cial lnc page tempor maxim tractabl subset allen interv journal associ comput page complet system polygon journal philosoph polygon dame journal formal page magnitud proc american conf ai page model integr qualit spatial dynam reason physic proc american conf ai page spatial reason object applic physic problem ieee conf ai framework dynam structur phd school comput lattic theori space comput mathemat page lehmann semant pergamon page transit challeng autom theorem proc cade model forc falt advanc qualit mit page cohn overview spatial logic base region principl knowledg proc intl conf morgan page metric properti physic levesqu reiter principl knowledg proc intl conf morgan page tractabl fragment region connect proc intl joint conf ai page canon model region connect schubert shapiro principl knowledg proc intl conf morgan page complex qualit spatial maxim tractabl fragment region connect arti cial page histor summari contemporari ieee comput graphic page ho constraint close comput graphic imag page theori qualit spatial reason base order proc american conf ai page semant natur case al semant movement work paper intl page visibl locat qualit singh qualit reason page frank kuhn spatial inform theori theoret basi proc lnc page shape frank proc specialist meet geograph object undetermin franci reason spatial arti cial page studi clarendon concav rectangular ieee transact page analysi guarino poli proc intl workshop formal ontolog conceptu analysi knowledg page essai hahn librari live open chicago la page page cohn overview qualit spatial visual proc intl joint conf ai page algebra structur set frank spatial inform theoret basi proc lnc page tarski geometri chapter clarendon review ai plan hendler tate read morgan san order theori approxim anger al proc workshop spatial tempor page prospect data knowledg page boundari smith white philosophi cognit proc intl wittgenstein de relat spatial phd paul reason stock spatial tempor page algorithm tempor proc american conf ai page region interest postal proc intl conf document analysi recognit de read qualit reason physic morgan san process correct free macmillan pub reason base giunchiglia selman principl knowledg proc int conf page axiomat method cambridg canon model class real spatial abel ooi advanc spatial intl symposium lnc page qualit spatial reason combin orient frank spatial inform theoret basi proc lnc page delta frank kuhn spatial inform theoret basi proc lnc page spatial reason concept motion sloman prospect arti cial io page 
Agents/1222 1 hierarch optim process wang sridhar mahadevan depart comput scienc engin michigan state east mi usa abstract gener strategi approxim solv larg markov process origin problem interact solv independ account natur interact paper focu class process nonstationari manufactur robot natur interact subtl studi compon state transit probabl depend polici solv interact subproblem approach solv parallel fail present approach variant explicitli account mode dynam merg algorithm combin base level polici present detail result transfer larg manufactur task show hierarch approach faster outperform heurist run transfer line factori introduct aspect autonom includ act studi framework markov process unlik classic method solv reinforc learn framework solv mdp prior knowledg underli state dynam reward rl agent learn polici act dynam uncertain environ process rl larg number success recent fundament problem current techniqu dont scale larger problem categor approach scale rl gener class approach construct simpler mdp defin tempor abstract polici primit action abstract approach address scale issu model call hierarch machin model call method gener strategi scale rl origin intract larg mdp exampl research decomposit approach dean lin al parr singh approach mdp decomposit reli specif characterist problem origin problem divid origin mdp state complet partial decoupl navig problem subprocess directli affect action set coupl smaller mdp previou studi typic approach success paper gener approach class process domain work manufactur practic import exampl found domain robot plai elev serv half floor natur interact subtl problem compon transit probabl depend polici solv interact defeat approach solv parallel analyz charact present hierarch approach approxim solv explicitli account natur interact approach base averag reward algorithm call smart solv variant constitu level approach combin base polici final polici solv problem section introduc framework section provid inform wai includ previou work decomposit studi section describ transfer line problem manufactur show characterist problem make suitabl method section present hierarch algorithm section present detail result hierarch method appli larg transfer line compar reinforc learn algorithm section summar paper process introduc framework process sketch reinforc learn algorithm solv assum underli knowledg transit probabl reward process defin tupl set set set state action depend transit reward function give probabl transit time pair gener mdp princip action histori depend model transit time distribut action xy denot probabl action system transit state transit probabl function describ transit epoch natur process describ state trajectori time function probabl epoch occur time action execut state comput denot probabl system state time unit action state epoch calcul expect transit time epoch reward function consist fix reward result action state addit reward accumul rate time natur process remain state epoch note natur process chang state time rate reward accumul epoch vari expect reward system state action chosen express oe transit time denot state natur process demonstr algorithm find polici algorithm estim action function sum reward receiv polici action subsequ stationari polici defin xy max tempor differ current state actual state updat action valu expect transit time replac actual transit expect reward replac actual reward action valu updat ff imm max imm actual cumul reward earn epoch due action state ae averag ff learn rate paramet note ae reward estim ratio total reward total simul time taxonomi method provid inform wai includ studi past relat problem class studi paper mention prior method mdp focu specif class mdp problem state space origin problem divid subset creat smaller subproblem paper broaden concept state space compon action set higher polici origin problem introduc concept term problem satisfi process divid pm notat ff abbrevi stochast approxim updat rule process subproblem action chosen fix polici note notion abstractli affect doesnt origin problem set subproblem util solut subproblem solut problem matter origin problem redefin state action transit probabl reward function subproblem tabl give exampl classifi natur interact navig tabl typic class problem suitabl state decomposit origin state space simpli divid result smaller subproblem coupl interfac mark tabl expect state optim polici origin problem solv respect polici construct hierarch polici case state intern compon action set state transit inherit valu origin remain problem solv process problem approach action coupl action air campaign plan problem mdp studi singh problem appli resourc task action influenc avail resourc task transit probabl action affect polici solv kind state action set constitu affect problem solvabl polici remain part view mdp special case state decomposit room room room room navig problem state part reward function room redefin reward compon remain unchang action decomposit task task action set task mt air campaign plan problem coupl set action composit mdp studi action set composit mdp proper subset cross product compon action space polici decomposit polici fix polici fix polici fix transfer line influenc transit probabl problem discuss detail section tabl wai base interact transit probabl reward chang chang polici make complic coupl requir approach examin coupl gener origin problem coupl case occur intern compon action set transit inherit valu origin coupl stronger compon subject chang due polici decomposit problem possibl constitut coupl section introduc real world instanc polici transfer line problem section discuss approach solv problem stochast model transfer line manufactur problem studi system consist serial combin distinct goal satisfi target demand minim inventori machin ag subject breakdown repair period mainten avoid unnecessari costli repair transfer line model factori assembl product buffer machin reduc effect breakdown machin product process intract analyt comput optim control polici transfer line machin heurist control method constant kei disadvantag heurist address problem minim inventori cost includ cost product repair prevent mainten cost factor wed optim solv process model analyz transfer line simplest singl machin fix complic case multipl machin singl machin transfer line introduc transfer briefli examin model singl machin figur singl machin action space set machin suppli materi upstream buffer work send downstream buffer raw materi suppli suppli finish product buffer remov demand rate action produc increment machin run accumul lead high failur machin repair repair mainten time stochast variabl model distribut suppli rate demand rate figur singl machin inventori model time action process suppli demand rate state space sm singl machin transfer line vector denot run time machin mainten denot inventori size machin time continu simplic paper discret interv turn state space finit reward function rm machin previou fix reward type action posit reward neg cost continu item store buffer failur repair result fix cost includ function equat fix demand rate suppli machin transit xy state action singl machin model kind suppli rate machin demand includ paramet configur machin failur repair produc buffer unit product repair failur final unit cost inventori buffer configur paramet constant singl multipl machin transfer line transfer line multipl machin transfer line model assembl line oper sequenc machin make finish part figur suppli rate demand rate product rate figur transfer line multipl machin view larg suppli demand rate input output action space union expand individu machin action mn mi label action machin machin natur machin state space transfer line bigger cross product individu machin state space entir transfer line state space larger cross product individu machin state complet intract process transfer line upstream machin product rate downstream machin suppli downstream machin product rate upstream machin demand rate transfer line figur machin factor machin upstream influenc machin product rate machin machin line transfer machin demand rate demand rate transfer line process smaller process sequenti coupl machin product rate demand rate suppli rate figur model transfer line set significantli smaller singl machin product rate machin machin machin oper fix make machin model suppli rate demand rate demand rate transfer line machin product rate directli affect neighbor machin intern dynam charact make strongli coupl studi characterist requir solv problem hierarch faster approach optim hierarch optim algorithm describ hierarch optim combin offlin reinforc learn algorithm solv variant constitu onlin merg algorithm dynam recombin learn polici origin problem coupl wai approxim solv problem individu simultan learn polici machin independ run copi reinforc learn algorithm refer algorithm individu machin index transfer process depend machin suppli rate demand depend product rate machin machin product rate machin upstream downstream process machin vari make learn individu machin polici transfer line extrem difficult slow process transfer line smart algorithm modifi method section vide sequenc process coupl actual product rate train machin part transfer train machin isol smart varieti boundari suppli rate part target demand rate part consum machin line machin train optim transfer line select action oper dynam polici algorithm summar figur algorithm bear similar previou work parr creat subtask combin make optim optim polici problem dean lin work polici comput base estim valu system assum problem divid simpli state imposs transfer line problem ffl reinforc learn dmax dmax run smart algorithm singl machin transfer line rate store obtain polici figur ffl dynam merg machin time interv observ product rate machin observ product rate machin set run polici time figur hierarch algorithm optim transfer line note level lower reinforc learn level learn control individu machin action higher level abstract dynam polici merg step base observ suppli demand rate machin upstream downstream refer techniqu result section present detail result hierarch smart algorithm optim transfer line compar perform algorithm standard industri compar benefit hierarch decomposit optim algorithm result paramet unit cost singl product inventori cost failur mainten cost reward receiv machin failur time model normal distribut standard deviat repair exponenti distribut time mainten exponenti distribut time produc product distribut approach depend distribut cost conduct experi cost distribut model optim line begin compar smart machin transfer line problem heurist method note machin train part transfer line figur compar averag reward averag inventori size polici learn smart produc heurist graph show result time unit inventori buffer size machin total inventori rate result averag separ run averag reward polici learn smart significantli achiev improv partli due fact heurist incorpor mainten polici smart number time unit averag reward smart number time unit avg inventori figur compar averag reward inventori size learn smart transfer line produc heurist minim repair cost polici learn smart achiev reduct inventori size hierarch optim larg line turn hierarch divid optim process offlin train phase onlin merg phase offlin train machin train isol rate suppli rate ran smart time unit result shown figur note perform support rate higher demand fill inventori space smart job balanc product rate suppli demand rate higher averag reward support rate singl avg reward demand rate support rate singl avg reward smart demand rate figur averag reward smart variant creat vari suppli rate polici onlin merg reinsert machin dynam merg algorithm figur comput polici transfer line machin oper select learn observ rate set machin track suppli demand rate event closest polici big transfer line figur compar averag reward polici learn algorithm versu averag run approach figur result singl run flat hierarch smart suppli rate experi show larger scale optim hierarch smart method result dramat improv approach support rate demand rate machin flat hierarch smart number time unit machin averag reward flat hierarch smart figur comparison averag reward versu machin transfer line solid dot mean outperform approach averag reward plot machin transfer line transfer line machin explor question happen transfer line machin break frequent machin treat algorithm present previou train bottleneck machin machin merg polici machin middl transfer failur time time unit figur plot averag run approach result train machin separ polici merg step appli consequ approach uniform polici special polici number time unit machin tl figur machin transfer line machin index ignor bottleneck machin result shown pictur train add time onlin merg phase vari tempor interv polici merg show vari time interv control polici merg step affect perform figur show averag reward polici learn transfer line valu figur show averag run time rang frequent observ rate estim configur machin perform number time unit averag reward time interv figur valu machin transfer line futur work paper algorithm solv larg origin simpler strongli interact compon transit depend polici solv gener algorithm optim instanc transfer line algorithm method solv variant individu account mode interact dynam polici merg strategi abstract level comput polici entir transfer line result show approach superior heurist oper transfer approach learn control machin parallel issu remain unexplor research exampl possibl challeng transfer line task agent learn environ plai soccer view agent agent train reliabl plai agent fix polici recent work emploi train session stone issu remain investig deeper abstract maxq taxonomi decomposit method intuit formal gener algorithm solv arbitrari form special step forward acknowledg research support part nsf career award grant michigan state grant refer review product control research product oper manag comparison control mechan accept intern journal product research elev perform reinforc neural inform process system sh techniqu plan stochast maxq method hierarch reinforc intern confer machin learn experi hierarch schedul polici simpl manufactur system ieee confer texa hierarch framework manufactur system ieee confer lo ca manufactur system engin prentic hall reinforc preliminari tenth intern confer machin simul model analysi reward reinforc empir machin factori simul averag reward reinforc fourteenth intern machin learn confer fisher morgan kaufmann larg weakli coupl markov confer uncertainti decomposit algorithm weakli coupl markov problem confer uncertainti learn hierarchi advanc neural inform process system model tempor abstract advanc neural inform process system colorado mit press markov discret stochast dynam program wilei seri probabl mathemat statist john wilei hierarch make stochast manufactur system bertseka learn dynam channel alloc cellular telephon neural inform process system singh cohn dynam merg markov nip pull altern intern journal product research learn system phd carnegi mellon decemb reinforc mit ma bradford book learn tempor abstract intern confer machin learn pp 
DB/911 3 repres queri xml incomplet inform serg abiteboul studi queri xml incomplet inform simpl model xml data simpl queri system incomplet inform spirit system develop imielinski relat scenario incomplet inform xml document continu enrich success queri document show system repres partial inform sourc document acquir success intellig answer queri impact complex enrich system queri languag addit featur result suggest approach achiev practic appeal balanc express tractabl research present motiv xylem project object data web xml document introduct xml data call inform collect web site store central fashion inform held complet due limit storag dynam natur web expir deal incomplet inform view incomplet repositori xml document continu enrich explor web queri crawl web document miss partial author support part author support part nation scienc foundat grant number work perform part visit inria permiss make digit hard copi part work person classroom grant fee provid copi made distribut profit commerci advantag copi bear notic full citat page copi post server redistribut requir prior specif permiss fee pod santa california usa copyright acm luc victor vianu uc san diego incomplet inform xml document satisfi de nition incomplet inform xml document enrich answer queri remot document queri pose action rst altern answer queri incomplet inform data answer complet repres answer describ incomplet knowledg sourc import variat answer provid piec inform hold similarli possibl hold similar mediat seek sourc addit inform fulli answer queri incomplet inform guid determin addit explor web sourc advantag data paper introduc incomplet studi increment mainten address issu answer queri pose incomplet quest simplic cienci main motiv factor choic led limit notabl limit queri languag base pattern match simpl select condit data simpli ed dtd ignor order compon assumpt xml element persist identi er framework captur broad rang situat practic interest exampl illustr incomplet inform natur partial xml tree repres data type inform style dtd repres data miss type miss data interest reminisc extens dtd includ rang data price special mechan de ne type element pend context call incomplet tree illustr incomplet tree exhibit partial inform miss natur repres xml document show simpli ed dtd sati ed input sequenc pair partial knowledg input repres incomplet tree maintain increment ptime queri incomplet set answer queri repres incomplet tree comput ptime show incomplet tree form strong system respect queri check ptime queri fulli answer data case data su provid ptime algorithm incomplet tree determin addit inform sourc order fulli answer set queri retriev inform incomplet tree increment maintain size grow exponenti sequenc pair discuss wai deal exponenti blowup extens incomplet tree call intuit add form altern size incomplet tree shown remain polynomi respect entir sequenc pair manipul handl incomplet inform exponenti check empti conjunct tree ptime regular incomplet tree altern exhibit restrict input dtd queri ensur incomplet tree remain polynomi sequenc pair manipul remain polynomi exhibit approach deal case incomplet tree grow larg practic rst approach consist small set addit queri chosen provid critic inform elimin unknown inform shrink incomplet tree prove queri chosen incomplet tree remain polynomi size respect entir sequenc pair input dtd approach heurist approach heurist gracefulli inform repres incomplet trade accuraci size incomplet tree show approach incomplet tree polynomi sequenc pair argu core model practic appeal start point deal incomplet inform xml model limit discuss impact extens show minor extens lead signi di culti system longer exhibit partial inform longer strong system queri answer inform high complex undecid extens concern queri extra featur includ option neg queri construct path data power restructur model transduc discuss extens framework persist id assumpt issu order paper organ section introduc formal model type system section deal acquisit incomplet approach exponenti blowup incomplet tree section discuss extens complex undecid result paper end relat work incomplet inform interest earli system focu search semant queri appli incomplet semant incomplet approach close world assumpt open world assumpt cwa state hold explicitli state incomplet state rule incomplet tree reconcil approach combin semant describ exibl miss state fact document data exist landmark paper laid formal groundwork incomplet null introduc notion strong system system spirit address tree model relat model benign form queri languag restrict relat algebra data complex handl incomplet studi program complex higher exponenti data complex rst note part studi null weak instanc updat incomplet inform receiv special attent scenario spirit data techniqu reminisc techniqu answer queri studi problem model base incomplet inform studi problem manag incomplet due data expir studi extens dtd special consid special mechan system closest work investig studi incomplet inform semistructur data framework result di erent formal framework present core framework incomplet inform de ne turn model xml document simpli ed system incomplet inform run catalog figur data tree formal model abstract xml document label tree abstract simpli es real xml document minor substanti model distinguish attribut distinct consid signi simpli cation tree xml document order discuss issu order section nite set nite set element set data valu denot element node data valu possibl superscript denot set label assum set data valu ration number set equip linear order simpli ed model xml de ned definit tree tripl nite root tree node label associ label node data assign node data tree denot apr data tree ht ing restrict node note pre xe tree share node tree type structur valid document dtd simpli ed version dtd call tree type tree type speci element set element children node label multipl constraint root restrict easili remov alphabet label auxiliari notion multipl atom describ children node label atom express ai distinct label symbol definit tree type alphabet tripl label call associ multipl atom call type satisfact tree type tree denot de ned obviou root label label wi number children label ai restrict wi child label wi child label wi child label ai wi set tree satisfi denot denot tree type understood tree type catalog catalog product product price cat pictur cat observ write tree type repres graphic catalog tree multipl edg figur queri de ne simpl queri languag select pre xe input tree claim su cient practic queri basic input tree depth start read node speci ed element possibl select condit data valu node involv pattern extract speci ed leav pattern node label call queri pre queri queri figur queri price subcategori electron product price queri pictur camera pictur catalog label tree condi root associ node label extend alphabet intern node label sibl intern node distinct label cond associ combin statement form neg label absenc node label children node pattern node adorn bar entir root node extract exampl queri shown figur formal notion answer queri auxiliari concept valuat condi input data tree map node node edg mi map edg label int sati es mi edg child label answer pre tree consist node imag valuat descend node label answer queri catalog depict figur remark insist essenti aspect model remark identi tree queri answer pre xe root share node input import aspect amount persist node identi er input answer condit tree type discuss incomplet inform main idea time result previou pre full data tree repres complet data queri initi type de nition partial knowledg miss portion full tree incomplet inform includ pre tree descript miss inform describ miss inform extend tree type disjunct multipl condit data ad special mechan de ne type element instanc figur pose queri miss data type rst product categori product price greater de ne formal incomplet extend notion tree type note data sourc continu tree de ned introduc de ne simpl condit tree type condit de ned queri simpl condit tree type alphabet pair map associ disjunct cond associ condit condit appli data node label set tree repres simpl condit tree type de ned obviou manner denot extend notat tree type condit tree type hand side product disjunct multipl ab found context dtd express structur properti depend context node consid special achiev type element suggest de nition special map map set element transform data tree data tree obviou replac label readi de ne complex type paper condit tree type special alphabet tripl simpl condit tree special map label special label catalog element product semant condit tree type de ned data tree exist tree similar condit tree type unrank tree automata de ne valid set role special alphabet condit tree type similar state tree automaton analog fulli lack presenc data valu tree avor automata techniqu carri set tree de nabl condit tree type properti similar regular tree languag set tree de ned condit tree type close complement technic point algorithm test empti set tree satisfi condit tree type easi reduct test empti grammar show lemma check empti incomplet tree discuss incomplet inform consist pre full data inform miss portion tree miss portion condit tree type alphabet special alphabet order fulli captur incomplet su cient provid label pre tree element provid interpret node term special alphabet readi de ne incomplet tree definit incomplet tree consist condit tree type special alphabet data tree td data label map associ node td subset includ denot incomplet tree catalog incomplet tree result queri repres figur grei node inform describ miss portion input tree catalog product price cat pictur catalog catalog product dtd sourc product product product canon elec elec soni elec product canon elec elec soni elec elec camera camera camera price cat pictur camera camera answer queri catalog product price cat pictur catalog product catalog product canon elec camera queri queri answer queri price cat pictur price cat pictur incomplet inform queri queri figur catalog product elec camera price set tree repres incomplet tree de nition consist tree exist data tree sati es pre node label symbol symbol label node set tree repres denot incomplet interest check fact set fact partial inform natur fact interest pre xe tree incomplet tree data tree pre tree pre node identi pre tree pre show standard techniqu dtd theorem incomplet tree data tree check ptime pre pre acquir incomplet inform present main result show framework cientli deal incomplet inform scenario rst deal acquir partial discuss exponenti blowup problem answer queri knowledg consist incomplet classic problem queri incomplet appli model issu knowledg order fulli answer queri acquir incomplet inform basic inform web acquir gradual answer queri show framework develop assum input singl document tree type case multipl sourc easili reduc case virtual merg sourc singl document input tree tree consecut answer ne partial inform describ incomplet tree stage incomplet tree answer ne incomplet inform comput describ tree compat answer queri nement algorithm call outlin inform algorithm ne rst illustr algorithm catalog figur incomplet tree queri data tree output queri incomplet tree describ miss product product return queri electron product price creat label obviou condit attach special product construct incomplet tree repres kind product return queri cheap camera pictur node id product canon answer queri inform return node queri merg note persist node id assumpt critic product queri queri type inform fact price product case type product queri queri case camera return queri categori camera camera pictur case infer type miss return queri queri electron camera pictur yield categori miss node note nement algorithm ne input algorithm incomplet tree answer consist condit tree type special alphabet data tree output incomplet tree comput condit tree type negat condit tree type tree return empti answer node queri tree creat type la label type special la intuit ta accept start assum involv empti ta accept start verifi condit accept start satisfi condit return empti answer queri propag empti set cond cond creat tree type process queri root leav induct root type tr root tq node children tq de nition children bound add rule ta accept accept ta condit sati ed ai children satisfi condit requir iter note node leav treat di erent propag eventu failur type leaf creat easi modifi step step consist comput intersect tree type input tree type straightforward polynomi time note yield tree type conjunct disjunct multipl atom produc condit tree type miss inform data tree obtain creat type comput join exist node id crucial data special comput intersect type data type special condit tree type comput show theorem incomplet tree answer algorithm ne comput polynomi time incomplet tree complex algorithm ne repeatedli increment ne initi tree type success pair increment step size incomplet tree exponenti sequenc illustr algorithm incomplet tree ptime complex incomplet tree exponenti respect sequenc pair algorithm develop worst exponenti complex respect sequenc legitim due system answer turn prove independ system illustr pre shown theorem ptime respect incomplet tree theorem type xed alphabet data tree determin pre ai determin pre tree node identi proof reduct avoid exponenti blowup exponenti blowup incomplet tree illustr tree type root queri root answer empti incomplet tree construct queri algorithm ne yield disjunct multipl statement avoid exponenti blowup incomplet conjunct disjunct type de nition incomplet restrict initi tree type rst discuss extens incomplet tree restrict initi tree type conjunct incomplet tree prevent exponenti blowup incomplet tree conjunct disjunct multipl atom type speci disjunct mean conjunct disjunct multipl atom tree simultan valid respect type speci ed conjunct term analog altern nondetermin control refer incomplet tree augment conjunct conjunct incomplet tree shown conjunct incomplet tree construct increment extens algorithm stai polynomi sequenc pair input tree type conjunct incomplet tree illustr incomplet inform provid pair repres conjunct root ai bi special discuss algorithm ne yield disjunct multipl dnf form conjunct price pai conjunct tree complex manipul kei problem check conjunct incomplet tree theorem polynomi usual incomplet tree lemma restrict tree type queri approach avoid blowup size incomplet tree restrict input tree type exhibit restrict reason practic input tree type restrict perform data test path queri tree condi linear test pair node ni parent straightforward modi cation algorithm ne produc input incomplet tree size polynomi sequenc refer variant algorithm ne show theorem tree type answer incomplet tree construct algorithm ne size polynomi proof queri conjunct node factor transform conjunct incomplet tree regular incomplet tree straightforward process exponenti maxim depth queri qi tree type depth qi depth bound algorithm remain polynomi qi heurist sketch approach deal case incomplet tree grow larg practic rst consist small set addit queri chosen provid critic inform elimin unknown inform shrink incomplet tree choic addit queri guid appli incomplet tree larg standard choic addit queri incomplet tree polynomi size larger complet input data tree tree type approach deal larg incomplet tree gracefulli inform order shrink incomplet tree idea illustr incomplet tree larg explicitli restrict pair valu test repeatedli queri make maintain inform connect valu costli inform connect retain rang valu combin valu identi ed score system maintain dynam queri approach extend combin type constraint data valu structur tree omit detail abstract queri incomplet tree show incomplet tree answer queri distinct issu rst classic problem answer queri inform provid incomplet tree answer incomplet provid descript answer rst issu address present section address incomplet tree guid mediat decid queri sourc document order provid complet answer user queri note import technic properti incomplet tree produc algorithm ne call incomplet tree reachabl construct algorithm ne initi tree type sequenc pair inspect shown reachabl tree special form ensur multipl statement incomplet tree involv special version symbol unambigu special statement assign data node possibl describ miss data impact properti complex manipul signi comput complement intersect condit tree type gener analog manipul reachabl incomplet tree polynomi time rst question mention knowledg world consist incomplet tree obtain algorithm ne mean data tree answer answer data tree incomplet tree form strong system set incomplet tree arbitrari theorem incomplet tree ectiv construct incomplet tree denot construct decid fulli answer inform provid incomplet tree corollari tree data tree decid ptime fulli answer proof reduc test empti condit tree ptime lemma remark import side corollari combin theorem provid check answer view provid sequenc pair problem answer queri view studi recent context relat import variant queri answer problem incomplet decid fact answer queri consequ theorem theorem reachabl incomplet tree data tree check ptime pre pre guid mediat partial inform input speci ed incomplet user pose queri virtual input document provid complet answer queri inform addit queri gener input document obtain inform fulli answer queri incomplet tree guid gener queri assum gener queri explor input document start node refer queri local order gener local queri determin sourc possibl inform relev queri state question ectiv theorem reachabl incomplet tree check tree tree proof similar theorem detail gener local queri show cientli comput set local queri collect addit inform answer incomplet tree data tree td local express form queri return answer full td root set local queri complet rel equal obtain extend node td queri pose root trivial complet rel point local queri avoid work previou queri complet properti node exist retriev queri node retriev distinct queri queri return empti answer input tree call arbitrari incomplet tree illustr incomplet tree queri incomplet tree queri complet rel queri appli retriev exist valid incomplet reachabl reachabl incomplet turn complet state theorem incomplet construct ptime set local queri form complet rel complet appeal enforc cost paramet account order gener cient complet extens framework xml document incomplet inform reli limit avail persist node lack simpl queri languag section discuss extens impact handl incomplet inform studi extens scope paper illustr kind di culti extens introduc de nition section inform begin discuss extens impact handl incomplet inform branch recal tree pattern child label node pattern disallow queri pattern simultan product pictur camera branch multipl children label incomplet tree remain strong system extend branch maintain increment ptime reachabl incomplet tree exponenti respect data tree special root root incomplet tree describ possibl assign valu branch construct answer queri construct answer consist bodi head bodi tree pattern node pattern label variabl bodi de ne set bind variabl input node head queri speci es construct answer data tree bind spirit skolem function incomplet tree longer strong system branch construct answer root root bodi head produc answer equal number bs bind bind incomplet tree exist strong system queri remain open branch option queri option label semant valuat partial requir de ned node option request camera displai pictur exist camera answer pictur combin branch option yield exponenti blowup complex question consid illustr variant pre question note theorem check ptime tree pre answer tree compat input tree type singl pair ai extend branch option show theorem tree type pair queri branch option pre proof reduct valid dnf formula disjunct note complex lower bound independ system extend transduc turn framework extend order tree power restructur long data join illustr tree transduc introduc model wide rang xml transform languag origin transduc work tree data valu easili extend select refer extend transduc acceptor analog call extend languag order data tree accept call extend regular tree languag input tree type de ned extend regular tree languag natur system queri extend automaton automaton data tree languag accept extend automata provid system maintain stai polynomi input type entir sequenc pair shown extend techniqu theorem input type speci ed extend pair qi extend transduc ai exist extend automaton comput ptime fact automata provid system broad class restructur drawback intuit appeal incomplet inform provid incomplet tree lost automata strong system discuss necessarili regular tree languag input type transduc problem occur simpler set branch simpl construct basic manipul handl incomplet inform high lower bound theorem extend automaton due thoma fact test empti gener regular express recal empti condit tree type test ptime lemma basic step manipul extens involv join data valu turn extrem power featur lead dramat di culti handl incomplet inform kei question undecid join data negat negat consist label express alphabet complement valuat match posit extens valuat match neg extend negat leav join data valu compar data valu di erent node pattern queri combin featur lead undecid question input tree type sequenc pair data undecid queri empti answer theorem input tree type sequenc pair queri qi extend data proof reduct undecid implic function depend easi shown undecid queri tree satisfi input tree type compat pair undecid tree pre tree note result independ system impli exist ectiv system empti pre decid join data option construct reduct similar proof theorem show theorem data tree input tree type pair queri qi extend data option construct pre answer queri tree theorem highlight option construct answer path express join data valu queri pattern node reachabl parent pattern tree path label spell word regular languag extend path express test data valu lead undecid kei question theorem input tree type pair queri qi extend path express test data proof reduct empti intersect languag show ectiv strong system class queri node id order inform discuss persist node id issu order node id signi assumpt framework avail persist distinct queri xml document return node id node ident avail node id enrich inform node consecut illustr catalog id longer gener system incomplet inform extend order track wai match node return di erent queri assumpt node id gener depend sourc provid persist node id gener id element approach expect process cost order issu order facet consid input tree answer queri preserv answer order element input sourc dtd describ order children possibl regular express edg queri order select pattern queri request element occur element condit regular weaker partial order condit discuss extend transduc show framework extend presenc albeit cost high complex clear order complic handl incomplet inform input element rst queri request list element order queri list element queri list element answer answer depend type input form answer answer answer input form answer previou inform interleav element problem similar issu persist id wrapper data sourc provid rank merg answer consecut queri absenc system maintain inform partial order element order issu interest question explor main contribut paper simpl framework queri xml document incomplet inform framework model xml document simpl xml queri system xml incomplet inform show incomplet inform acquir consecut queri answer cientli repres increment ned system queri handl cientli exibl answer provid incomplet answer system full answer provid complet partial inform addit queri framework limit wai assum sourc provid persist node id order document dtd queri queri languag path express data join order trace boundari consid extens framework show signi impact handl incomplet rang high complex undecid justi es featur make suggest practic appeal solut handl incomplet inform xml refer abiteboul duschka answer queri materi view proc acm page grahn queri set world theoret comput regular tree languag alphabet unpublish beeri milo schema integr translat structur data proc int conf page shim optim queri materi view proc icd page mediat data proc acm page codd understand relat bull acm sigmod page complex relat queri inf yang expir data proc page grahn problem incomplet inform relat berlin yannakaki test instanc assumpt inf proc imielinski incomplet inform relat sagiv queri incomplet answer semistructur data proc acm page widom prototyp data creation mainten proc acm page sagiv answer queri view proc acm page yannakaki complex test implic function join depend vianu typecheck xml transform proc acm page papakonstantin vianu dtd infer view xml data proc acm page ullman answer queri templat bind pattern proc acm page reiter sound complet queri evalu algorithm relat null valu person complex problem automata theori logic phd report mac project mac vardi complex relat queri languag proc acm page vardi integr incomplet inform proc acm page zaniolo relat null valu journal comput system 
Agents/147 1 research council canada institut inform technolog nation de recherch canada institut de de manag system atm network yang phan juli ieee intern confer network franc juli nrc copyright nation research council canada permiss grant quot short excerpt reproduc figur tabl provid sourc materi fulli acknowledg manag system atm network yang phan nation research council canada abstract report paper develop manag system atm network pm path manag solut requir manag complex atm network aim assist network oper perform oper simplifi procedur automat optimum rout select aim provid effect support fault identif prevent network oper introduct atm commun network plai import role telecommun network wide backbon transmiss access network emerg larg heterogen atm network challeng research develop area network manag manag atm commun network dramat size virtual manag consid import task task consist includ path path upgrad path fault fault fault qo servic manag exist cot softwar compon provid atm switch vendor provid partial support oper perform procedur oper dispar difficulti oper perform automat oper manag assist oper perform automat oper manag complex heterogen atm develop manag system atm network base proactiv framework framework gener framework facilit incorpor technolog develop network manag system develop environ solari platform microsoft peer nt platform pm offer simpl mechan set perman virtual path connect perman virtual channel connect user interfac establish automat manual requir qo paramet throughput delai monitor maintain manag perform traffic data fundament alarm incid associ pm detect problem futur perform correct prevent automat feasibl situat automat action pm notifi network oper detail inform natur locat reason procedur correct prevent develop pm atm network manag tool servic provid network oper effect manag network provid good qualiti improv network perform reduc loss research develop main oper support automat optimum rout select fault identif prevent pm system architectur implement oper manag present paper section discuss system section describ oper section experi final section conclud paper system architectur pm develop base architectur architectur accommod exist manag protocol standard corba underli distribut middlewar vp oper support idl devic applic vp monitor agent fault agent snmp servic agent devic decisionmak devic configur agent perform agent corba bu corba bu commun agent dmi devic figur architectur view tier servic tier data tier kei characterist architectur separ distribut comput environ data compon build maintain applic corba chosen distribut middlewar stabl standard matur product object request broker invok method remot object necessarili locat exact function corba client manag distribut devic explicit knowledg topolog network shown top tier view layer middl tier servic layer bottom tier data layer develop tier implement view tier tier made user interfac applic applic invok method object middl locat locat platform pm view tier user graphic interfac oper access control atm network remot locat browser download applic consist user oper network network element statu support fault knowledg base updat servic tier servic agent main servic agent configur fault perform agent monitor agent perform monitor manag task includ fundament function recommend osi provid servic request view tier applic agent distribut locat design support corba idl commun agent perform oper manag simplifi oper automat optimum path fault identif fault predict automat fault correct prevent data tier tier made object interact manag system pm develop manag atm network equip tier made commun protocol object dmi commun agent perform interact oper devic agent cope request servic agent devic agent viewpoint map protocol data object idl data object deliv servic agent detail map data idl protocol report paper oper manag goal pm assist network oper perform oper simplifi procedur automat path rout provid effect support fault identif prevent describ briefli fault fault prevent oper oper configur princip manag task atm network dispar proprietari manag procedur provid individu atm switch vendor creat environ difficult autom process situat increasingli futur network continu develop intricaci volatil solv boyer al configur intellig mobil agent approach requir atm switch provid environ run mobil agent network exist atm switch provid snmp support remot access distribut agent access atm switch sequenti send request switch request corba client idl data packag servic configur agent determin oper sequenc pack oper scenario pass agent oper agent determin action command perform oper support commun protocol methodolog shown system perform delet creation set path set atm switch atm upgrad rout path negoti paramet exist delet path set atm switch atm network oper pm snmp request snmp request idl request idl switch switch switch figur methodolog oper object provid oper oper goal provid simplifi procedur assist oper perform oper oper worri switch oper locat node connect switch procedur oper hidden system goal assist oper perform oper support automat path rout path select oper determin path rout thousand paramet network topolog reason determin path rout select optim path request path oper detail automat rout select path report paper due space limit fault identif fault identif prevent main manag task pm reduc human error misunderstand process fault approach consid effect approach number achiev appli approach alarm correl alarm correl system reduc amount detect problem viewpoint manag oper make fault identif fault prevent oper make fault correct prevent assist oper effect manag develop system identifi problem make fault correct system implement knowledg base consist pm system gener manag atm specif knowledg shown fig pc po defin fault problem stand path hardwar path softwar path path incid set defin problem fault identif main infer data fault fault fault correct data collect fault disord occur hardwar softwar manag atm network traffic densiti path rout alarm event extern manifest fault alarm event defin atm vendor gener atm equip observ network oper alarm event import data collect alarm event detect problem order effect detect path congest path collect traffic data monitor reason traffic data alarm event lost real time manag system notif collect alarm event manag atm switch traffic data monitor fault detect task fault detect find symptom problem identif collect alarm event traffic data definit fault detect gener incid fundament alarm correl traffic data associ incid open incid set problem descript alarm traffic data analysi incid associ dm collect data traffic data analysi alarm incid incid detect incid associ conflict figur fault identif knowledg isol alarm correl network manag alarm correl aid oper network fault reduc amount alarm reduc amount fundament alarm correl gener incid incid associ fundament mechan alarm correl tempor boolean relat traffic data analysi task traffic data analysi collect traffic data gener incid traffic data analysi detect congest overload problem monitor bandwidth manag bandwidth obtain collect traffic data bandwidth time close base fault ph ps correct decisionmak support automat oper greater bandwidth incid formul attribut time stamp locat assign group equip current statu set congest sever assign open incid incid associ incid associ task incid associ open incid set problem term gener incid alarm correl traffic data analysi system classifi group obtain incid incid set map incid set monitor fault isol incid set open fault confirm problem task fault isol isol problem open incid set incid confirm support mean loop test assumpt monitor problem time fault correct fault correct determin effect reactiv action identifi problem task fault correct strategi problem correct give support inform assist oper fault prevent order prevent fault provid proactiv manag predict futur statu manag atm network dynam network adopt neural network approach predict futur statu histor data current statu reflect dynam chang atm configur data includ neural network train input paramet train design neural network path nn compon predict futur statu monitor path predict fault detect procedur detect futur problem problem proactiv control action prevent problem detail fault prevent refer refer experi test prototyp system evalu effect support capabl fault identif built experi network consist atm network simul small real atm network commun protocol pm atm network simul corba pm real atm network snmp atm network simul gener raw alarm event raw traffic data term standard specif alarm event gener simul network problem traffic data gener term qo paramet manag simul network problem real atm network set fore system switch atm adapt instal sun workstat set sun workstat atm switch hard test effect support capabl fault identif prevent base prototyp manag system atm network incorpor corba technolog pm provid effect support network oper path manag automat rout simplifi oper fault identif proactiv fault predict prevent futur enhanc knowledg base manag focu automat fault correct prevent servic manag refer yang identif prevent manag atm atm june pp gabriel robert mark expert system shell telecommun network alarm ieee network manag control pp gabriel mark ieee novemb pp robert gardner david alarm correl network fault resolut kohonen ieee nov pp expert system fault diagnosi italian telecommun integr network pp edward network diagnost expert system multiplex base gener network diagnost integr network pp elliot expert system diagnosi advanc digit subscrib pp atm network architectur base virtual ieee transact vol pp yu watanab dynam rout scheme intern ieee commun vol pp inter domain symposium integr network pp john bernard toni configur heterogen atm environ intellig mobil st intern workshop mobil agent telecommun oct pp 
DB/2173 3 version manag xml document chien vassili tsotra carlo zaniolo abstract problem ensur effici storag fast retriev structur document import recent popular xml document semistructur inform web tradit document version control model document sequenc line text shortest edit script repres version ineffici preserv logic structur origin document approach structur document preserv timestamp hierarch effici reconstruct current past version call base copi control gear effici version reconstruct small storag overhead analysi experi illustr effect approach version control structur document easili support multipl concurr version partial document retriev introduct problem manag multipl version xml semistructur document interest content provid cooper work xml standard problem transport level webdav work group develop standard extens http support version meta data space overwrit protect view problem queri effici multipl version adapt techniqu tempor persist object manag flat structur tradit file storag suit support demand complex comput queri content document restructur interest manag structur document tag object timestamp version number lifespan rest paper organ review state art version manag limit current method deal xml document method section comput scienc ucla depart comput scienc uc riversid research partial support nsf depart comput scienc ucla overcom perform evalu section section discuss gener manag multipl concurr version support cooper author effici retriev document occur section interest identifi tabl content keyword index previou work tradit method version rc scc effici structur document method store version edit inform differ requir reconstruct singl version result cost comput complex reconstruct document object rc store current version intact store edit script script describ backward document develop histori version current extra process appli edit script gener version popular text version manag tool scc append version differ end scc interleav edit oper origin code associ pair timestamp text segment lifespan segment version retriev scc file scan file retriev valid segment base timestamp rc scc read extra text segment invalid retriev result addit process cost total cost proport size current version size retriev version current version situat version file read version retriev rc reduc version retriev cost maintain index valid segment problem valid document segment store page gener version creat page miss featur rc scc xml document preserv logic structur origin make comput document difficult support copi control control version retriev copi store version document object strategi total cost linear size target size total number current target version cluster document object notion page page simplic assum document evolut creat version linear version version impli version establish appli number delet latest version document object repres record object id object attribut lifespan interv delet insert version fill version object ad document object delet version physic delet version object record updat version delet place formul concentr object insert updat object version repres delet object insert updat object version artifici insert creat record share oid subsequ lifespan interv initi assum object document version physic store page logic order number object specif version physic scatter disk page page store object version retriev specif page access complet object access page invalid target version assum version document consist object record store data page size object page evolv histori version object object object version object delet defin full page version percentag page correspond valid object page version fall version object delet version updat invalid record record store page full final page fall influenc object version cluster page high impli object version store fewer version reconstruct access fewer page page version maintain minimum page version minimum perform paramet page fall minimum valid record page copi page similar oper tempor index reconstruct version reduc access page fast page good part request version space remain linear number document version histori defin refer full page assum data record written page singl page full record extend definit includ page default affect perform version page scheme rc scheme perform version minim document chang version reconstruct version requir retriev overhead rc perform poorli grow larger version retriev version cost time retriev store complet version cost overhead retriev version limit storag overhead situat rang minor major mix histori document adapt method oper case tend store complet timestamp copi scheme achiev desir behavior merg rc scheme copi control scheme rc problem solv rc store edit script object size script reconstruct time approach separ document object edit script cluster document object edit script page reduc overhead long version histori summar scheme figur show version document version store shown figur version store page assum size document ch sec sec ch sec sec sec ch sec sec data page assum maintain minimum page page threshold version default version version creat sec sec sec sec ch ch sec ch sec sec size sec ch sec sec data page logic order object version ch sec ch sec sec ch sec ch sec sec sec appli version version version fig sampl xml document version page part version page origin sec sec part page remain chang affect page useless version valid object copi data page copi object includ ch sec ch sec sec determin object copi object insert page logic order shown figur object copi object store page base linear order version edit script reconstruct record edit script version script version shown figur deriv origin edit script copi object treat delet oper insert oper delet oper ad ch insert oper attach oper posit target object version posit sec version notic posit delet object posit version delet posit sec version delet oper posit posit valu recov total order object mean discuss detail version gener sec sec sec sec sec sec sec version data page ch sec sec ch sec sec sec ch sec sec version data page ch sec sec ch sec ch sec sec sec version data page sec sec sec ch sec sec sec fig version file sec sec sec sec sec size sec sec sec sec sec page page valid object ch sec copi object copi object store data page logic order version version shown figur insert algorithm illustr figur auxiliari hash tabl improv version insert process hash tabl store valid object current version data page object hash tabl page current version version reconstruct discuss retriev object store data page gener version object store logic order comput page base version edit find page insert object copi object data page base order gener script obj diff replac updat consecut oper copi object insert insert oper delet oper diff logic associ posit oper diff object posit posit version object posit object front associ insert oper page number page target store script updat hash tabl fig version insert algorithm step reconstruct logic order object logic order recov fashion base script sampl version figur explain algorithm describ reconstruct version reconstruct start retriev object version find object edit oper gap oper posit miss object fill gap previou version start retriev object version retriev start gap retriev object previou version find object version return version object back version version receiv read data page root ch output page main memori valid sec version reconstruct version continu previou stop object object current oper mean target object delet previou version version request object version expect sec answer request version retriev object retriev previou run similar version request object int count edit gap request script previou version append return object list count count posit posit current gap posit current oper target script previou version append return object list count count oper script previou version retriev object posit posit current append object current oper return fig version retriev algorithm version oper return record delet oper object version retriev request issu version version version back sec object version version send record back version repli request expect insert record record search object version continu check edit oper find object posit procedur continu script object version retriev edit script snapshot algorithm reconstruct version version file reconstruct version involv script data page version read request version reconstruct page total number version size total edit script accumul sooner affect version retriev effici control overhead read edit size edit script reconstruct version threshold size edit script snapshot built version edit script snapshot insert record object version edit script figur edit script snapshot gener edit script snapshot prevent version edit script earlier version complex analysi reconstruct data page version read version total order version object recov edit script result edit oper list page point pointer record read sequenti object retriev page useless page read valid object page copi page object version store object store page gener version page version version object gener version store order version retriev process merg order object list merg object list involv page instanc process mean memori hold page version read retriev process discuss applic retriev version cost reconstruct version page read edit script read page version size version number denot capac page snapshot page number page creat version bound requir size determin object copi object object includ insert object updat object delet object remov delet affect size denot total number document evolut updat object part bound copi object number object copi bound object copi copi object copi total number object copi bound number object copi time bound total number copi object bound total number version snapshot rc total number version snapshot rc fig version retriev storag cost requir document total number version snapshot rc total number version snapshot rc fig version retriev storag cost document size evolut total number copi object combin size linear perform evalu evalu perform scheme examin version retriev cost total size compar perform rc scheme copi space requir compar snapshot scheme simpli store document snapshot version snapshot scheme minim version retriev cost version directli retriev disk page size set byte examin behavior document size remain approxim version total number version snapshot rc total number version snapshot rc fig version retriev storag cost document size evolut set experi size document version page version previou version insert half uniformli randomli distribut document data page document evolut consist version figur present result examin scheme snapshot scheme lowest version retriev cost document size retriev cost proport version actual cost depend paramet lower impli page invalid object version object scatter data retriev time observ figur largest retriev graph show rc strategi read prior target version retriev version earlier version cost retriev version version size snapshot scheme minim version retriev storag cost prohibit quadrat total number rc minim storag cost store space case grow linearli number rate storag overhead scheme due copi object higher page useless faster copi made experi design explor evolut document size expand shrink expand evolut version document size approxim version insert perform result figur version retriev cost scheme proport document size actual overhead depend paramet storag cost linear size total note space cost close minim storag rc small delet percentag rare copi useless page figur illustr perform document size shrink rate version consist delet insert version retriev cost document shrink rc scheme retriev cost grow linearli number result observ perform scheme tune paramet space close rc retriev time approach optim snapshot scheme actual choic paramet set base applic characterist requir retriev gener multipl concurr version paper concentr linear sequenc version document version valid time notat denot sixth version document version number replac time interv version valid softwar engin common model version version stem call minor origin branch call major branch point branch evolv independ instanc denot minor branch stem version major branch minor branch stem case author start paper parallel branch subdivid branch cooper case branch merg case multipl author part document merg final paper document process repres direct acycl graph version model significantli power origin model captur document stage evolut model verif progress baselin exploratori support verif comparison individu effort situat richer model add complex origin scheme scheme separ edit script minor branch strictli script share part major script simplifi assum branch start fresh edit script snapshot minor branch current page comput assum object minor branch aliv rest includ part consid expir aliv object minor branch copi page object minor branch written object major branch written current page point separ branch manag object branch assign page page comput view empti part current page minor complic script involv manag current page time minor branch gener minor major branch write current fill mixtur fail test branch copi current page made major branch minor branch avoid unnecessari copi major writer current branch point minor decid current page assum empti space major page copi object object branch merg branch accomplish simpli branch continu object edit script continu branch simpli refer object branch percentag object copi page branch fall standard copi oper perform structur index document version previou section problem retriev complet version document user frequent interest retriev part section version document consult tabl content document user chapter section version support partial edit script augment marker facilit identif major logic structur document object retriev desir section desir version retriev pertin portion edit page desir segment identifi script similar problem occur common situat index version document entri index point object document keyword case manual case direct index succe retriev actual find difficult decid version section belong point directli index point structur edit actual object inform version number section easili obtain solut made edit script snapshot scheme rc scheme gener solut index problem reconstruct version section chapter cost reconstruct singl section point tabl content index reconstruct document nest list model add logic structur marker edit denot start end separ list element entri edit instanc marker posit denot linear locat edit script version figur structur sec sec sec sec edit script object uniqu identifi logic posit sec refer retriev sec search structur edit script structur edit document chapter section easili answer issu request retriev document segment posit posit edit script augment version denot maintain structur edit insert object record posit structur script marker insert delet edit script updat structur document retriev part segment chapter section structur edit script snapshot version built standard version object order reconstruct shown figur obtain edit script ch sec ch ch sec sec reconstruct posit object check rang data page object read object data page read reconstruct process stop reach end requir rang reconstruct complet sec read index structur built top structur edit invert index assum entri keyword ad index built version medic document object kei word entri record index index entri denot keyword object segment version keyword version document segment queri version retriev chapter mention answer check invert index locat target document rang locat rang version structur edit script unnecessari version unwant document segment read altern approach index entri point actual easi reconstruct logic structur version gener scheme simpl flexibl approach implement index version document document structur document repres specif xml tag gener tree structur yield portabl gener tabl content keyword index control manag evolv structur document effici requir minim extra space compar tradit strategi strategi improv effici version retriev cost control copi investig perform altern techniqu store includ extens persist list method tree method tempor comparison show method perform approach term storag retriev effici shown method flexibl platform support kind document queri concurr version index document structur rc page list method discuss provid flexibl investig relat includ queri restructur version effici gener browser output transport internet version document includ automat gener dtd version document dtd document refer optim access symposium larg spatial vol pp parallel softwar vj manag xml document webdb vj compar studi version manag scheme xml submit icd wagner manag acm softwar engin symposium practic softwar develop environ access method ieee transact knowledg data vol pp lomet access method acm sigmod marc sourc code control ieee transact softwar dec pp fast algorithm unit cost edit distanc tree journal salzberg vj comparison access method acm comput vol walter system version experi juli pp vj snapshot access method inform intern vol robert wagner michael correct problem journal vol januari pp www distribut author version 
IR/522 5 automat discoveri languag model text jami margaret du center intellig inform retriev comput scienc depart massachusett massachusett usa abstract text larg organ internet make difficult person search languag model describ content select algorithm gloss provid assist automat select inform current practic languag model cooper approach import limit paper demonstr cooper requir select servic construct languag model sampl content normal process run queri retriev document random sampl approxim carefulli select queri sampl approach avoid limit cooper enabl addit capabl result demonstr accur languag model learn small number queri document introduct text larg organ internet make difficult search desir inform larg corpor network provid access thousand lotu note oracl corpor document manag text document access thousand searchabl text person inform select algorithm gloss solut permiss make copi part work person classroom grant fee provid copi made distribut profit commerci copyright titl public date notic copi permiss copi post server redistribut requir prior specif permiss fee sigmod usa fl acm select algorithm select automat rank text likelihood satisfi queri select algorithm requir manual scale effici million handl inform appeal solut respect select algorithm interact directli rank algorithm interact index languag model describ content content languag model vari select gener languag model describ word index term occur frequenc inform indic term occur languag model essenti element select research activ direct studi concern paper interest select servic acquir languag model provid languag model request call cooper assum parti provid inform parti cooper model turn weak solut environ manag parti paper present method acquir languag model requir special cooper sampl assum select servic run simpl queri retriev small number document languag model built automat document sampl test seri experi result demonstr accur languag model learn small number queri document sampl avoid problem cooper approach enabl addit capabl support cooper method section describ select includ languag model acquisit approach languag model section methodolog result section section discuss sampl enabl addit capabl relat section conclud select peopl commerci search servic face select problem peopl search select subset simplest choic search choic practic environ commerci servic group set common newspap collect court group manual process automat method common group effect solut inform inform access pattern anticip inform select effect solut select algorithm rank similar queri queri yield rank tailor queri typic top threshold similar select search select algorithm appeal handl rang moder comput requir compar ordinari full text inform retriev effect demonstr research studi effect languag model prohibit comput commun cost compar queri select algorithm compar queri index partial repres content call partial languag model languag model arbitrarili complex tend consist list term occur frequenc occurr complex languag model includ inform phrase term type inform unclear acquir languag model start protocol techniqu acquir languag model start protocol al consid iso standard interfac inform retriev system start protocol exchang languag model protocol basic idea simpl list index inform frequenc occurr small amount corpu inform number document suffix strip word extrem frequent word remov start protocol excel solut singl parti control reli make poor solut control parti fail provid wont lie cooper system upgrad inaccess system reli cooper solut start cooper protocol enabl provid cooper select servic provid hostil select base corpor simpli incent cooper select servic uncommon inform provid internet order number peopl visit site start protocol offer protect problem problem assumpt vocabulari frequenc inform provid compar difficult occurr compar occurr occurr ir system varieti word stem stopword special index common date select algorithm find imposs differ system start protocol ideal suit environ control singl cooper index uniformli characterist small organ larg approach requir acquir languag model sampl select servic recommend search find document satisfi inform observ impli capabl run queri return document match queri minim criterion assum satisfi queri run document sampl content document return queri necessarili constitut call techniqu sampl characterist larg popul estim small random sampl popul hypothesi sampl provid sampl document suffici random learn accur languag model entir open question larg sampl requir construct languag model accuraci word occurr follow highli skew word occur word occur rare word middl frequenc rang thought distinguish document singl evid highli frequent word distinguish bit evid suggest import vocabulari occur frequent acquir sampl resourc measur queri run document reason algorithm sampl simpl select initi queri term run queri retriev top document return updat languag model base characterist retriev document assum necessarili run queri return document free cost acquir languag model issu li scope paper stop criterion reach select queri step algorithm involv number specif queri term document examin stop sampl defer discuss choic section paper sampl number advantag techniqu learn languag model requir special cooper run queri retriev assum minimum level servic requir absenc special cooper enabl sampl learn languag model wont cooper make content languag model learn consequ normal behavior build languag model directli sampl document select servic control content languag model free select servic imposs task reconcil approach text index stopword case select servic determin sophist detail appli creat languag enforc consist languag match detail languag model characterist select algorithm document sampl dn rich resourc enabl select servic provid addit capabl select enabl complex analysi content drive visual aid union sampl sampl serv select enabl set queri import piec inform difficult acquir sampl size zipf law empir evid show vocabulari growth addit document rate independ size unclear estim size sampl size primarili select algorithm scale word frequenc languag model provid vari size similar effect obtain scale frequenc learn languag model size sampl base make size byte document uniqu term total term varieti cacm homogen heterogen tabl test corpora piec desir methodolog hypothesi motiv work accur languag model learn sampl text simpl queri hypothesi test compar languag model learn sampl languag actual languag model record number queri document requir learn languag model achiev level accuraci languag model experi conduct languag model consist index term frequenc frequenc measur number document term frequenc stopword discard languag model construct control learn actual languag model compar word languag effect discard learn languag model word consid stopword default stopword list inqueri ir frequent word suffix remov word languag model construct control suffix remov prior comparison actual languag actual languag model stem choic consist languag model explor research literatur select test effect corpu characterist size quickli accur languag model learn homogen set titl abstract scientif articl commun wall street corpu american newspap heterogen consist trec cd newspap magazin scientif govern document standard test corpora research tabl summar characterist metric languag model consist type frequenc inform term vocabulari correspond learn vocabulari actual vocabulari measur percentag learn ratio correspond learn frequenc inform actual frequenc inform measur spearman rank correl coeffici metric percentag learn learn vocabulari necessarili subset actual learn examin subset document natur proport set cover proport term actual vocabulari found learn vocabulari call metric percentag learn percentag learn metric poor match text skew distribut term text uniqu term text occur occur occur time vocabulari text word occur time percentag learn metric treat term languag model equal import realiti convei inform content ratio skew distribut term text larg number essenti irrelev small number frequent term frequent term convei inform content metric measur qualiti learn vocabulari weight import term ratio metric measur proport term occurr cover term learn languag model learn vocabulari actual vocabulari ratio number time term occur term ratio mean learn languag model word account word occurr consist occurr occurr learn languag model ratio spearman rank correl compon languag model term frequenc rel import descript power term inform retriev algorithm typic frequenc inform compon statist rank procedur term occur consid repres content sampl algorithm estim proport document term estim size imposs estim actual number document term estim proport document term compar actual proport document comparison number document examin true proport accur estim document amount error built type comparison accur rank term frequenc occurr compar rank term occur learn languag model zipf law predict relationship term rank frequenc term frequenc estim vice versa spearman rank correl coeffici accept metric compar rank spearman rank correl coeffici defin rank differ common term number term rank ident correl coeffici uncorrel coeffici order coeffici select requir rank correl coeffici suffici learn languag model repres rel import index term suffici rank term select algorithm insensit small rank open question error algorithm toler select accuraci deterior li scope paper paper simpli studi correl learn actual rank varieti condit set paramet experi sampl requir make choic queri term select document examin queri queri run determin select word randomli actual languag model initi queri select select frequent select term languag model inform experi found choic initi queri term effect qualiti languag model learn subsequ queri term chosen varieti section case term chosen subject order avoid select term retriev document term select queri term number requir charact long requir similar requir index term text retriev system guid document sampl queri experi conduct determin effect vari paramet cacm experi present paper end examin document experi present paper end document stop criteria chosen empir run initi interest learn languag model small size sampl experi continu thousand document ensur unusu happen cacm number document examin percentag term found cacm number document examin ratio figur measur learn languag model cover vocabulari percentag term cover learn languag model percentag word occurr cover term learn languag model document examin result seri experi conduct method gener queri number document examin queri vari goal experi determin effect languag model learn combin paramet produc fastest learn accur languag model queri run determin select word randomli baselin experi consist subsequ queri term randomli languag model examin document queri number determin paper present result number document result baselin experi shown figur graph figur demonstr percentag term learn poor metric judg qualiti languag model document order discov cacm vocabulari repres word occurr cacm contrast dramat vocabulari discov repres term occurr recal stop includ frequent term discard languag model comparison stop word word languag syntax convei inform stop word discard ratio grow converg rapidli cacm number document examin spearman rank correl figur measur term rank document frequenc learn languag model term rank document frequenc document examin percentag term found examin number document function vari wide examin ratio experi frequent term found examin fix number matter big ratio greater curv impli point term occur rare consist zipf law figur show rate rank term learn languag model begin match rank term actual languag model model small homogen cacm converg correl coeffici exce model heterogen converg reach correl coeffici document heterogen converg reach correl coeffici document unlik converg constant rate correl term rank influenc size model cacm reach high sampl model reach lesser sampl model sampl experi demonstr repres term frequenc learn sampl small fraction content question answer rank error distribut evenli opportun gather inform proper rank frequent term rare term measur distribut error subject current research open problem correl rank accur select correl coeffici requir correl coeffici suffici sampl continu reach level correl requir sampl queri greatli improv size sampl number document examin queri baselin experi retriev highli rank document queri number baselin produc good result empir number paramet reason investig effect vari cost run provid process build languag model wait cost reduc examin document queri searchabl web return document titl addit document request characterist suggest examin document queri cacm doc doc doc tabl effect vari number document examin queri long sampl method reach ratio doc number document examin spearman rank correl coeffici process run queri retriev document intend approxim random sampl queri necessarili return sampl document document return queri similar vocabulari term frequenc gain examin suggest examin small number document queri seri experi perform determin effect vari number document examin queri valu test difficult graph differ accuraci languag model learn examin number document includ paper differ sampl produc stabl result rang paramet set point ratio reach languag model homogen cacm learn quickli document examin queri document queri larger heterogen document queri larger heterogen make differ document examin queri differ suffici small base rel cost run queri examin document choic requir fewer queri reach number result show cost examin document sampl number document examin ratio number document examin spearman rank correl figur measur queri select strategi affect accuraci learn languag model percentag word occurr cover term learn languag model spearman rank correl coeffici term rank learn languag model wall street journal document examin avg number queri tabl number queri requir retriev document queri select criteria queri select strategi baselin experi select queri term randomli learn languag model random select learn languag model baselin paper produc good result empir queri term select learn languag model select languag model queri select strategi reason investig effect vari earli hypothesi frequent term produc random sampl occur varieti context metric measur term frequenc common inform retriev document frequenc collect term frequenc averag term frequenc tf metric differ tend test method select queri term learn languag model earli concern learn languag model initi strongli document happen sampl bia reinforc continu select queri term learn languag model solut select term complet languag hope random set queri term hypothesi design languag hypothesi compar languag techniqu previou set experi languag model experi full languag model choic creat unfair bia favor interest languag model actual languag model provid investig carefulli seri experi methodolog previou set experi number document examin queri queri term select learn languag model term frequenc metric randomli languag model test result similar result present tabl random languag experi learn import vocabulari term inform quickli random languag requir queri creat sampl document differ select queri term dont occur sampl occur fewer document result learn languag model similar languag model exact match sampl superset view result number fail queri higher select queri term languag model similar experi demonstr select queri term learn languag complet languag produc markedli skew sampl document rate learn faster measur number queri slower measur number document examin whichev metric languag model learn moder cost experi demonstr select queri term randomli learn languag model effect select base high frequenc result hypothesi high frequenc term occur weak produc random sampl hypothesi valid experi high frequenc term tend frequent similar wall street journal care approach select high frequenc base part speech pai attent relationship sampl produc result simpli random select queri term produc random sampl document select stop criteria term rank learn languag model converg term rank actual languag model number document examin differ learn actual rank long disappear complet stop criterion sampl system decid learn languag model suffici accur hypothesi learn languag model converg actual languag model rate hypothesi support extent experi spearman rank correl coeffici initi rise level rate converg slow predict inform stop criterion simpl techniqu compar learn languag model time learn languag model time ffi learn languag model suffici conclud learn languag model converg similar languag model determin spearman rank correl semant obviou intuit experi show identifi small improv correl languag model learn document interv tend highli correl coeffici notic differ correl actual languag model defin measur averag rank differ term rank view averag measur percentag number term move convert rank defin rank differ common term number term rank term ident term rank rank term rank rank term swap term occupi vari multipl term occupi case languag vari figur show languag model creat increment languag model creat cacm document languag model creat cacm document mean averag term rank move distanc reach place number term rank common valu observ gener indic languag model chang dramat document increment interest result valu snapshot languag model fell document independ size cacm number document examin chang document figur averag distanc term move convert rank document rank document document examin result suggest establish stop criterion base inform observ languag model learn languag model accur consecut document span fall independ relat discrep ratio spearman rank correl coeffici languag model converg ratio suggest converg constant number document spearman coeffici suggest rate converg influenc size ratio constant number document suffici experi suggest stop condit creat depend inform observ learn languag model question stop condit suffici sampl constant number irrespect size characterist peak summar content interest primarili automat method learn languag model suffici accur detail automat select algorithm languag model person gener natur text simpl method summar content displai term occur frequent stopword method effect word occur illustr point world base microsoft custom support web chose content present late base earliest sampl unclear paramet set sampl experi document examin queri subsequ research fewer document queri effici term learn languag model rank avg tf rank produc metric easili document microsoft softwar avg tf produc inform word rank list content word draw strong summar content test simpl word list effect summar necessarili effect techniqu frequent common relationship word consequ creat languag model sampl document make power summar sampl process restrict word list frequenc restrict inform provid set hundr document mine frequent interest inform inform enabl construct power inform summari simpl languag model cooper method queri queri process term ad queri document make longer repres person inform queri occur user assist interact document retriev queri base term word phrase tend occur queri necessarili synonym phrase alien ad queri tend document term avg tf term avg tf term avg tf term avg tf term avg tf project microsoft access set command excel object print applic offic user data product window work visual internet menu server beta error text set word servic box softwar tabl basic articl code version printer file setup messag nt mail system inform field user dialog select tabl top word found sampl microsoft custom support avg recent paper show queri significantli improv automat select arguabl import languag model inform document term occurr occur document occurr document polit queri algorithm requir larg repres document analyz pattern ordinari document search inform task clear expand queri select queri specif introduc bia select open problem queri gener select task sampl method build languag model solv problem build languag acquir sampl dn union sampl frequenc frequenc pattern occur set serv select algorithm union sampl favor specif reflect pattern common queri select view advanc make queri common part automat select peopl inform retriev system decad ago ir system million peopl form internet system inform comput network store commerci scatter thousand searchabl text manag varieti larg small content provid challeng research commun make inform access inform singl todai automat select solut incomplet solut environ internet cooper natur languag model dissemin protocol start address varieti problem encount world environ sampl approach languag model acquisit present paper avoid problem cooper protocol appli older appli system specif incent easili defeat intent avoid problem reconcil word custom index text problem arguabl problem cooper appli provid intend cooper paper show sampl produc accur languag model text vari size content hundr document document acquir run hundr queri resourc measur amount amount network low addit benefit sampl document directli select servic construct set languag model vari complex simpl word frequenc languag model select detail languag model identifi relationship select servic document acquir sampl build cooccurr queri type queri significantli improv accuraci open problem acquir paper present solut work report paper import step direct automat step select algorithm number document scale open problem size estim sampl content criteria recogn languag model suffici accur principl criterion subject current research queri select criteria lead rapid learn languag model acknowledg paper base work support part nation scienc librari congress depart commerc cooper agreement patent trademark offic advanc research project arpa order issu contract recommend express paper necessarili reflect sponsor refer croft search distribut collect infer network eighteenth annual intern acm sigir confer research develop inform page acm distribut scalabl mechan distribut inform retriev fourteenth annual intern confer research develop inform page octob acm dumai latent semant index text retriev confer page nation institut standard special public jc jc cl kj prei evalu select testb experi annual intern acm sigir confer research develop inform retriev garc paepck start stanford internet intern confer manag gravano garc gener gloss broker hierarchi intern confer larg page garc effect gloss text discoveri problem intern confer manag page editor text retriev confer nation institut standard technolog special public heap inform comput theoret aspect academ hp luhn automat creation literatur abstract ibm journal marcu comparison effect comput human search intermediari journal american societi inform mj editor fact figur nation inform standard organ inform retriev applic servic definit protocol specif em nk learn collect fusion strategi eighteenth annual intern acm sigir confer research develop inform page acm xu callan effect retriev distribut collect annual intern acm sigir confer research develop inform page acm zipf human behavior principl introduct human ecolog 
Agents/2522 1 factori control albert baker work field factori control decad span work factori control schedul group gener motor technic control system laboratori ge corpor research develop hewlett fellow renssela polytechn institut center manufactur faculti member electr comput engin comput scienc depart cincinnati dr baker develop success implement schedul system part doctor dissert develop demonstr factori schedul system rock island armi darpa project fund intellig autom continu work area implement simul agent robot weld person assist agent dr baker work summer faculti scholar unit state air forc manufactur technolog treasur offic cincinnati design test manufactur engin ge major applianc busi earn renssela polytechn harvard rice studi result contract net manufactur novemb pp case studi agent bid actual cost schedul paradigm distribut resourc ed manufactur control contract simul intern confer comput integr pp floor schedul control simul shop control confer march pp system plan schedul manufactur european journal oper vol pp zweben intellig schedul san morgan alloc distribut factori ieee vol februari pp research inventori review oper vol pp inventori product inventori englewood prentic pp design factori futur studi toyota product system industri engin edit product jone machin chang world person person factori control schedul flexibl manufactur ieee robot autom schedul manufactur neural journal neural network spring pp job shop comput industri vol pp jone neural genet algorithm oper research product plan ed pp neural network flexibl manufactur system comput industri vol pp lee network applic schedul job parallel comput industri vol pp constraint satisfact problem neural ieee intern confer neural san june featur map travel salesman neural vol pp wilson stabil travel salesman problem algorithm hopfield biolog vol pp foo neural network solv ieee intern confer neural san pp foo linear program neural network ieee intern confer neural san pp xu neural algorithm travel salesman neural vol pp krogh introduct theori neural redwood publish pp yamamoto minim total actual flow time neural intern journal product vol pp hopfield comput optim biolog vol pp tank optim signal linear program ieee transact circuit vol pp neural network intern joint confer neural vol pp neural network approach ieee transact neural vol pp convent neural network heurist job shop spie vol scienc artifici neural pp schedul american product inventori control societi confer pp de product inventori control ed pp race north river theori constraint north river process ongo improv great north river control contract comput system engin depart phd renssela polytechn factori control analysi heurist travel salesman algorithm ed academ bell system technic vol novemb pp perform bound simpl pack siam journal vol pp bin pack doctor depart massachusett institut survei techniqu discret annal discret vol pp lagrangian relax method solv integ program manag vol januari pp lagrangian relax base schedul algorithm semiconductor test fourth intern confer comput integr manufactur autom octob pp manufactur system lagrangian relax ieee transact automat vol juli pp plan schedul robot assembl fourth intern confer comput integr manufactur autom octob pp ramaswami control autom manufactur intern seminar manufactur ann simul part ii color number oper vol pp simul part oper vol pp cd gelatt mp simul vol pp evolutionari algorithm commun vol march pp jr algorithm appli assembl spie confer kb ai system aerospac spie code number algorithm job shop comput industri vol pp shop schedul genet intern confer genet algorithm pp genet algorithm job ieee intern robot april pp genet algorithm product vol august pp problem travel genet edg recombin intern confer genet algorithm san pp neural network expert system fm confer flexibl manufactur pp hybrid artifici neural network expert system approach flexibl manufactur system ph factori control shortest commun vol pp shortest rout symposium graph pp note problem connect numer vol pp matrix graph siam journal vol pp ghosh algorithm shortest iee vol pt march pp algorithm singl sourc shortest path vol pp ritter analog parallel distribut shortest path ieee intern confer comput vlsi comput septemb pp levitt arrai solut graph commun vol pp parallel depart comput cornel technic report tr parallel shortest path vol pp minimum delai rout algorithm distribut ieee transact vol januari pp garei applic multiprocessor siam journal vol februari pp combinatori complex comput miller ed plenum pp lawrenc product plan mathemat program applic industri engin manag optim mathemat program model techniqu practic john wilei son optim model system south san scientif polynomi algorithm linear vol pp english translat soviet math vol pp algorithm linear vol pp posit integr solut linear american mathemat vol pp relat siam journal vol pp davi product schedul stochast manufactur intern journal comput integr vol pp theori ed englewood prentic pp reduc oper vol pp lund hard approxim minim journal associ comput vol septemb pp papadimitri complex journal comput system vol pp factori control task alloc discret manufactur paradigm distribut resourc ed lin rout control confer flexibl manufactur juli amsterdam pp lin flexibl rout intern journal flexibl manufactur vol june pp lin shop floor control autonom vol juli pp opportunist part west school industri purdu august cooper control model ed nist special public pp garei comput guid theori freeman schedul oper vol pp gonzalez complex oper vol pp johnson flow shop job shop mathemat oper vol pp network afip spring joint comput vol afip press pp deadlin loss manag vol pp tree parallel schedul ieee transact vol march pp schedul oper vol pp product line minim maximum ucla technic report manag scienc research project report lo encyclopedia comput scienc vol ed marcel pp job machin sequenc algorithm minim number late manag vol pp algorithm sequenc job minim total annal discret vol pp optim naval research logist vol pp stage product schedul setup time naval research logist vol pp extens johnson result naval research logist vol septemb pp horowitz approxim algorithm schedul journal associ comput vol april pp jr minim flow ifip congress publish compani pp appli englewood shortest distanc graph improv induct discret vol pp factori control baker ieee vol decemb pp shop stack modern machin vol septemb pp floor control lead survei american institut scienc confer pp survei schedul oper vol pp jh phillip survei dispatch rule manufactur job shop intern journal product vol pp theori schedul publish han control maximum throughput rate manufactur materi handl research georgia tech technic report februari han time workflow control integr multistag manufactur materi handl research georgia tech technic report februari shaw bid distribut plan flexibl ieee confer artifici intellig miami dec pp schedul cellular manufactur framework network journal manufactur vol pp shaw plan cellular flexibl manufactur manag inform research center technic purdu west design distribut approach intellig manufactur inform phd west purdu plan schedul approach comput integr nb symposium optim autom manufactur januari pp distribut schedul method comput integr local area network cellular intern journal product vol pp shaw adapt distribut artifici intellig workshop distribut artifici lake control flexibl manufactur intern confer intellig manufactur june pp control manufactur journal manufactur vol pp control flexibl manufactur robot vol pp control heterogen manufactur system journal manufactur vol pp oper larg comput manufactur west industri engin purdu decemb commun control manufactur control ieee intern confer robot pp comput commun cambridg august factori control jr schedul tempor aiaa comput aerospac octob cooper approach larg scale product american associ artifici intellig workshop product plan st august ow schedul opportunist annal oper vol pp smith cooper schedul intern confer expert system lead edg product plan ed pp schedul princip autonom ed ann septemb nation center manufactur scienc ann cooper manufactur intern seminar manufactur pp cooper manufactur robot vol pp evolutionari ieee system vol august pp distribut schedul manufactur journal manufactur vol pp mitchel comput architectur ieee vol pp intent system issu ecolog psycholog vol pp shortcut vol pp dynam collect ant intern confer simul adapt meyer septemb mit press pp differenti wasp model selforgan group intern confer simul adapt meyer septemb mit press pp distribut behavior comput vol pp flock vol pp simpl solut pursuit eleventh intern workshop distribut artifici februari pp manufactur cell control dynam north american manufactur research confer pp holon manufactur system home ed fractal revolut corpor cultur intern organ ottawa report refer model manufactur iso version nation research council unit manufactur process research unit manufactur issu opportun research nation academi person petersen motion multiag appli artifici vol pp factori control net dog gen agent creat web world report pp secret internet advisor pp intellig privat internet pp wave product hit internet world pp model design englewood prentic softwar approach negoti singl function agent base design concurr engin research vol pp balasubramanian intellig design system integr manufactur intern confer san pp distribut mobil robot autonom decentr robot ieee intern confer robot pp product plan control process manufactur work depart manufactur boston technic system part blackboard model problem solv evolut blackboard ai vol pp system part blackboard applic blackboard system knowledg engin ai vol pp flow intellig intern confer manufactur scienc technolog june flow orchestr distribut manufactur west purdu august flow orchestr intern confer product august pp methodolog architectur reactiv intellig zweben ed san morgan kaufmann pp collabor implement blackboard technolog shop partner expect ann nation center manufactur septemb pp ghost machin macmillan architectur heurist factori american control june pp experi contract distribut artifici ed lo morgan kaufmann pp actor distribut manufactur ieee confer artifici intellig dec pp burk distribut asynchron system predict reactiv artifici intellig vol pp burk distribut asynchron intellig zweben ed san morgan pp jr tempor relat task hierarchi job shop intern confer comput integr pp factori control necessarili algorithm implement agent architectur develop lack area paper address question algorithm inspir case contract net form forward backward schedul call forward backward continuum schedul inspir common agent concept bid current work develop determinist simul form determinist simul attempt autom human interact requir success implement system gener open question algorithm agent architectur impli perform algorithm distribut comput worthwhil investig commun overhead algorithm evalu gener report research assur algorithm implement implement requir excess commun overhead refer im intern collabor japan industri robot associ nation center manufactur partner expect ann goldman centuri manufactur view agil competitor virtual strategi enrich custom van nostrand agil forg frontier virtual social ieee vol pp genesereth commun vol pp ken lang pai market wall street journal pp lot wall street journal pp blame retail web slow start wall street journal pp softwar consum send make web wall street journal pp race build intellig time pp factori control produc straightforward put agent stage product agent hold number specif lot size number unit lot size agent actuat product stage part bring stage inventori upstream agent actuat product pull algorithm straightforward implement pull algorithm popular earli work simul agent execut pull algorithm report flavor technolog alli signal implement pull base agent architectur factori summari paper start factori control algorithm implement addit research requir implement commonli algorithm factori control algorithm answer question major technic issu dispatch pull implement technic issu implement forward backward algorithm architectur common algorithm industri implement activ area continu research make determinist simul algorithm fit agent architectur research fit simul genet algorithm success algorithm central system report schedul consid subfield work report make standard techniqu work blackboard function heterogen agent architectur normal commun technic issu continu implement neuro schedul optim schedul algorithm gener make remain open issu implement algorithm agent architectur algorithm factori control review pull algorithm pull algorithm year gener thought inventori manag techniqu factori control algorithm toyota motor compani pull algorithm factori control chang gener percept toyota pull algorithm base simpl card assembl stage product inventori complet sit inventori batch lot size lot stage product produc exist stage fill lot stage assembl line stop begin lot pull final assembl final assembl lot back final assembl final assembl author produc lot finish product order make final pull lot previou return stage product stage pull lot stage return stage process continu line stage produc advantag pull algorithm schedul dispatch algorithm document pull inher inventori manag advantag identifi problem assembl line fix problem pull algorithm requir avail inventori pull system algorithm larg number manufactur make product small batch toyota success system great deal excit pull algorithm claim schedul vendor custom backward schedul algorithm pull algorithm ration schedul algorithm work backward custom job due date determin schedul reason common claim market mismatch excit pull algorithm applic algorithm market claim mean term manufactur terminolog term pull origin restrict deplet inventori signal product distinct import backward schedul true pull algorithm common algorithm perform perspect pull pull algorithm essenti lend implement multiag classic pull algorithm toyota implement manual shop floor personnel personnel requir central system factori control da success aspect make realist schedul real factori machin disappoint backtrack constraint propag techniqu requir hierarchi agent da make correct method optim schedul attempt appli standard intellig schedul techniqu provid cort cort techniqu solv schedul problem system implic cort great algorithm lead develop system cort agent schedul set job set resourc resourc share share resourc coordin resourc monitor agent constraint oper perform resourc deriv graph heavili demand resourc expect time agent send partial demand graph resourc mother produc final demand graph view agent oper resourc high demand time consid critic oper agent attempt schedul critic oper nomin time requir resourc lowest demand author leav open possibl schedul time prefer oper demand graph modifi constraint propag relat oper agent agent conflict agent move schedul oper request agent move oper cort system pursu special problem resolv content agent minim work version cort requir agent schedul resourc lowest expect demand polici result poorer schedul agent schedul time prefer matter expect resourc demand number research blackboard agent architectur solv schedul problem system appropri fall categori schedul system interest reader refer paper pull algorithm pull algorithm deplet inventori signal manufactur product inventori deplet deduc inventori reach level period review inventori level common tradit pull algorithm polici inventori fall econom order item order product algorithm call pull algorithm product initi product pull inventori factori control intellig schedul intellig schedul artifici intellig techniqu solv schedul problem techniqu method develop artifici intellig commun schedul uniqu compar algorithm present paper standard artifici intellig concept blackboard knowledg expert system number techniqu method develop artifici intellig commun specif solv schedul problem constraint conflict iter constraint conflict partit artifici intellig commun schedul initi isi project fall field numer failur success field report success number commerci product introduc intellig schedul techniqu standard techniqu fit realm heurist algorithm defin attempt standard intellig schedul heurist agent limit success intellig schedul expect substanti success techniqu method cast attempt made da asynchron system burk da implement report da architectur consist type knowledg constraint mainten system knowledg sourc knowledg aggreg process plan call agent origin develop agent repres resourc initi formul stabil problem algorithm implement final formul agent aggreg resourc agent schedul process final architectur arrang hierarch assign oper perform assign oper respect schedul oper respect resourc agent oper assign schedul oper list oper heurist schedul assign constraint propag system constraint mainten system determin oper schedul overlap time report conflict manag backtrack reassign relat oper report conflict resolv conflict order oper schedul relax constraint job factori control schedul job basi time bucket continuum time current factor incorpor prototyp capabl consist basic technolog involv hierarch schedul system ow al call system cooper schedul system variant origin concept eventu instal develop shop pratt aircraft css system work order schedul time agent call success oper work order schedul forward pass request bid agent repres resourc factori manag award tent contract oper perform issu request bid oper work order oper tent manag plan complet time manag similar bid scheme backward schedul complet time minim hierarch structur set balanc physic scalabl architectur determinist simul determinist simul schedul techniqu origin develop associ factor product idea run singl simul factori dispatch rule simul predict happen factori schedul term distinguish simul tradit stochast simul techniqu multipl simul run statist collect analyz expect system perform determinist simul schedul perform great deal human human schedul run simul number dispatch rule schedul modifi select dispatch rule resourc dispatch rule move job explicitli sequenc oper shop floor human schedul abil adapt chang determinist simul schedul refer capabl advanc run dispatch determinist simul lend implement develop determinist simul schedul implement work replac human comput interact autom tune schedul instal wisconsin factori control short week larger time comput resourc determin schedul amount time add compon product determin multipl expect process call factor schedul system oper time process time complet oper schedul system flexibl factor resourc factor chang base shop factor type job forward backward schedul oper fix plan week quarter schedul chang real factori oper forward backward schedul system differenti chang schedul accommod forward backward schedul system period weekend schedul total chang mondai morn call schedul system portion schedul effect chang factori floor modifi perform batch continu baker contract net provid success implement schedul system system implement perform studi simul ge job shop agent architectur perform type forward backward schedul bid compon resourc provid finish product custom negoti directli agent agent resourc perform finish touch product request bid compon agent resourc make compon request bid bid term actual versu custom select cost desir order issu network agent receiv reserv capac meet product bid bid show cost deliv desir time includ cost deliv earliest forward time backward time contract net determin cost continuum schedul forward backward schedul tradit forward backward schedul accomplish form contract net forward schedul quot backward schedul time slot quot contract net perform forward backward factori control ii time figur time phase product structur raw materi manufactur manufactur schedul mean manufactur resourc expect finit schedul techniqu finit schedul infinit perform finit schedul place finit backward schedul infinit backward work back ad compon product requir resourc addit compon slip back time resourc previou oper branch slip back time finit forward schedul similar time compon addit slip forward time util resourc encount finit schedul order schedul result schedul finit schedul job order discuss oper order oper highest prioriti oper oper backward forward schedul oper ad priorit list highest prioriti oper schedul resourc variat oper schedul prioriti oper determin resourc oper schedul resourc prioriti order expect util schedul call form schedul implement schedul conform theori constraint resourc schedul multipl resourc add specif compon product assur resourc util schedul finit schedul accomplish infinit schedul heurist method move job time slot resourc util schedul vari depend size time slot schedul time slot call schedul time bucket factori control ii end item assembl assembl figur bill materi late fact heurist design favor specif cost metric heurist rang optim design favorit cost metric compar heurist heurist method commonli techniqu industri method solut reason amount matter larg problem schedul common form schedul erp system determinist simul recent innov quickli gain broad industri intellig schedul artifici intellig techniqu factori schedul recent report level industri success success implement schedul system report heurist schedul schedul schedul schedul common form schedul erp system commonli schedul techniqu manufactur concept forward backward schedul pictur figur show bill materi product bill materi list time requir creat compon time phase product structur figur backward schedul complet finish product put product due date time compon ad product backward portion forward schedul earliest branch put time job system compon product ad schedul base rel locat forward backward schedul call infinit mean schedul develop assum infinit capac infinit schedul necessarili feasibl infinit schedul manufactur resourc schedul hour work hour dai infinit schedul factori control network learn dispatch rule work factori situat neural network combin determinist simul simul discuss section network factori dispatch learn situat paramet work complex dispatch rule neural network approach solv schedul problem gener suffer lack difficulti find good solut realist size inabl meet problem constraint realist size problem scale problem standard model network consist node number oper schedul formul unmanag number job system formul difficult implement scalabl base real entiti manufactur system job machin neural network formul lend zhou formul analog neural network node oper agent architectur agent oper model quadrat energi cost zhou linear cost function emploi nonlinear process feedback connect result number node number interconnect grow linearli total number oper approach appar requir research formul grow gracefulli small problem machin job larger problem machin job attempt realist size problem report network solv minimum averag problem formul extens schedul problem minimum extens demonstr paramet neural network energi function depend significantli size problem obviou obtain paramet trial error analog neural network formul continu commun provid node network clear commun limit prohibit actual implement agent connect digit commun network heurist schedul method heurist schedul techniqu bring good proven rang optim evolv optim solut cost metric set heurist serv motiv desir reduc tardi motiv heurist job factori control probabl exponenti approach depend long algorithm run algorithm simul anneal system brittl temperatur system high atom portion schedul jump spot local prevent system immedi local chanc global optimum found impli simul anneal algorithm put machin agent randomli assign job work agent sell job machin work agent split save earli arrang agent made cost agent cost anneal temperatur probabl enter arrang final schedul temperatur deriv time consid global central construct actual implement system interest research project genet algorithm genet algorithm base schedul number initi schedul meet constraint system schedul evalu give rank order schedul schedul mate mate portion schedul join form schedul good schedul mate good schedul bad schedul mate bad crossov schedul spawn care attent system constraint met tricki schedul evalu worst schedul thrown algorithm repeat number iter result longer improv time limit reach genet algorithm multipl schedul solut schedul agent claim genet algorithm number problem claim problem central mate algorithm distribut problem lack physic correspond scalabl growth clear put genet algorithm form outlin section neural network factori schedul neural network solv schedul problem neural factori control job lagrang multipli correspond type machin architectur solv lagrangian relax problem mixtur job agent recent project lagrangian relax techniqu agent architectur type agent al architectur type agent oper product part coordin agent system develop product process plan schedul product schedul process perform interact product coordin agent coordin agent perform task decid step size steepest ascent determin lagrang multipli agent violat goal distribut physic correspond outlin earlier function central construct simpl appar distribut desir ramaswami joshi approach solv integ program agent architectur suppli lagrang multipli result central comput agent job agent base happen shop agent modifi multipli job agent chang job rout determin lagrangian relax happen shop job agent end schedul determin lagrangian relax expect schedul follow close origin schedul lagrangian relax techniqu techniqu factori schedul gener agent schedul current work area begin work perform pursu concept fulli optim schedul number schedul algorithm attempt gain optim attempt optim cost metric algorithm wai chang valu improv optim schedul techniqu includ simul genet neuro schedul simul anneal simul anneal algorithm increment valu wai meet constraint valu factori control approxim algorithm minimum parallel machin schedul problem preemption limit case optim algorithm show result obviou wai implement algorithm rule algorithm solv minimum parallel machin schedul problem preemption sort easili meaningfulli implement agent system johnson algorithm bin pack problem implement algorithm sort job size alloc order bin room job job alloc bin algorithm requir distribut sort implement agent architectur agent repres job bin uniqu type optim algorithm recent proven factori schedul optim bound lagrangian relax techniqu lagrang multipli move constraint constrain optim problem cost function integ program formul equat constraint put cost function result optim problem optim sl subject lagrang multipli end function variabl variabl end function lagrang multipli optim problem solv iter solv solv call solv problem constraint variabl integr relax term optimum bound optim solut dual problem step optim lower bound optim algorithm continu iter reach small number iter reach fix number lagrangian relax techniqu provid result bound lend implement common formul subproblem equat factori control optim algorithm provid result close bound optim distinguish paper optim bound algorithm optim bound algorithm find solut absolut constant lower upper bound optim solut optim algorithm probabilist natur includ numer probabilist queu network simul provid solut probabl distribut complet amount mont carlo probabilist algorithm provid probabl optim complet amount time la vega probabilist algorithm provid optim solut defin complet probabilist provid optim solut expect probabl distribut optim schedul worthwhil result area part success theoret practic perspect problem optim problem optim problem optim algorithm exist measur larg make algorithm useless weight longest process time sort rule optim algorithm knapsack problem implement sort algorithm problem mean answer optim answer case arguabl approxim agent implement obviou horowitz provid approxim scheme problem minim job parallel uniform preemption provid scheme case machin scheme minim weight averag appar algorithm solv travel salesperson problem triangl inequ horowitz approxim scheme knapsack problem al ln travel salesperson problem triangl inequ travel salesperson problem distanc citi obei triangl inequ citi repres distanc citi travel salesperson problem remain longer factori control variabl restrict result integ program standard integ program formul factori schedul variabl number job manufactur resourc period prefer term hour longer reduc size problem solut job resourc time discuss gener factori schedul problem gener consid integ program clear integ program solv attempt decomposit davi jone integ program spatial group singl central comput coordin result subproblem solut hierarch architectur decomposit level resourc decomposit job entiti lend gener integ program problem research regard recent attempt approxim solv integ program agent architectur discuss optim schedul realist size optim schedul problem make redefin problem solv attempt redefinit relax requir solut optim requir solut rang optim optim rang optim defin close solut optim close solut bound optim probabilist descript type optim algorithm optim algorithm provid result close optim absolut approxim algorithm find solut absolut constant optim approxim scheme find solut fraction optim algorithm find solut fix fraction optim algorithm find solut fraction optim depend size problem factori control expect restrict case result polynomi formul research polynomi formul implement mathemat program mathemat program problem class problem mathemat techniqu make program meant program comput program system gener mathemat program formul equat easili solv mathemat program linear program linear program formul solv polynomi algorithm linear variabl restrict real cost function constraint linear function variabl linear program formul written optim subject ax xr linear program solv special case gener factori schedul problem determin produc produc assum product continu quantiti liquid mix linear program decid expend resourc cut stock problem rout formul problem solv linear program plan formul decid period expend resourc human demand period inventori carri cost period period solut direct implic job factori job gener factori schedul problem formul integ variabl restrict integ optim subject factori control mn processor solv problem agent mn relationship job machin architectur lack physic correspond desir simpl optim schedul problem solv polynomi distribut algorithm implement bruno algorithm minim averag parallel machin number problem relationship schedul problem implic schedul problem solv robert simpl schedul problem pose problem shortest path problem find shortest path graph connect node connect weight problem find shortest path pair node solv time central number node graph simpl matter arriv parallel shortest path algorithm solv time node node iter node directli connect node reach cost reach node iter requir comput node shortest path reach node node send work order follow path shortest path problem solv time node agent problem solv faster time polynomi number processor clear processor algorithm implement number agent grow number machin system interest variant shortest path problem provid consid case weight path amount traffic path case weight correspond delai queu assum amount traffic enter system stochast limit algorithm converg node shortest path node algorithm develop implement node lend agent implement problem similar schedul problem includ travel salesman problem bin pack problem knapsack problem problem agent implement factori control problem metric refer singl machin maximum late tardi singl machin number tardi job singl machin averag tardi singl machin averag ident parallel preemption weight averag flow time flow shop machin job shop tabl simpl optim schedul problem solv job agent sort time balanc distribut comput physic correspond agent job system algorithm scale standard central sort algorithm superlinear order ln simpl optim schedul problem solv distribut algorithm distribut directli map agent architectur rule central algorithm solv minimum problem job parallel machin preemption rule solv problem process step algorithm shown rule processor make distribut algorithm put algorithm job cluster group job group requir agent algorithm map directli physic correspond architectur question distribut balanc scalabl reason problemat map solv minimum averag problem parallel uniform machin preemption problem solv central algorithm ln problem solv time machin ident job process time machin uniform job process time proport constant depend machin job process factori control schedul algorithm differenti optim sought algorithm paper discuss schedul algorithm optim schedul simpl mathemat program optim optim heurist schedul optim schedul optim schedul problem formul gener optim problem constraint problem select paramet cost function maxim paramet restrict set subject constraint paramet formul boundari condit succinctli written optim subject optim schedul problem shown solut evalu determin solut gener mean problem solv reason amount fastest comput growth number solut number job manufactur resourc problem solv reason amount time matter architectur optim schedul problem includ minimum problem parallel machin preemption minimum problem minimum minimum minimum tardi problem simpl optim schedul problem gener believ realist size optim schedul problem section discuss simpl optim schedul problem tractabl simpl optim schedul problem simpl optim schedul problem solv tabl list tractabl optim schedul problem solv distribut sort algorithm case job system agent agent perform distribut factori control dispatch rout base earliest finish time rule detail analysi util compet machin process speed scenario analyz bid congest system time produc bid larg lin consid agent part agent negoti machin agent currenc dispatch rout price part agent pai resourc agent charg vari depend market condit develop simul system show dynam price set mechan stabl ting dispatch agent repres part product network manag dispatch algorithm easili implement rule requir local inform make straight forward distribut system scale addit agent effect dispatch agent case rout great deal agent research agent make collect bid potenti machin job rout content bid simpl inform earliest start earliest finish time case number bid system grow linearli number candid machin system consid grace scale architectur number machin grow larg bid congest occur practic filter put system make machin win bid make bid schedul algorithm schedul algorithm determin advanc job factori resourc term paper advanc schedul factori substanti schedul provid number advantag dispatch schedul system reduc random reduc inventori discuss surround figur schedul search global resourc optim optim result schedul manag bottleneck make machin bottleneck schedul custom order plan equip inventori plan interact gener tool inventori schedul add level rigid manufactur system advantag disadvantag factori control process time achiev advanc schedul fcf dispatch rule result geometr inventori similar result achiev standard dispatch rule queu model dispatch dispatch rule local inform specif manufactur resourc job make factori control lend implement agent research manufactur develop agent architectur implement dispatch rule common dispatch rout assum sequenc determin shaw agent factori control phd thesi manufactur cell subcontract work cell bid mechan cell finish job earliest finish time put criteria bid select suggest dispatch rule make rout shaw architectur physic correspond cell machin individu machin implement manufactur system hierarch system work report parunak al bid mechan mean agent find favor agent work rout associ report wisconsin term implement resourc materi handl repres agent part repres agent part finish process manufactur part request bid resourc perform subsequ process select resourc rout commerci dispatch system instal flavor technolog gener motor truck bu paint facil distribut paint booth repres agent agent decid paint job booth dispatch job highest prioriti random select dispatch rule make rout sequenc research consid featur extens dispatch agent architectur provid complementari factori control normal averag averag process util balanc figur fcf dispatch machin advantag dispatch rule simplic tempor predict futur make base present rule factori extrem dispatch rule spatial requir inform locat implement dispatch rule simpl algorithm requir comput easi novic understand implement expect behavior common dispatch rule dispatch factori easili analyz simul tool case analyz mathemat queu theori stochast model queu model show potenti difficulti dispatch factori queu network model show resourc util approach inventori dispatch factori geometr approach infin basic result random system figur show theoret result singl machin factori job arriv randomli independ exponenti job random amount time process exponenti job sequenc dispatch rule arriv process time determinist random inventori geometr approach infin util arriv perfectli balanc factori control terminolog unnecessarili import distinguish dispatch advanc schedul perform system control dispatch rule substanti system control advanc schedul factori suit dispatch control substanti suit schedul control review dispatch algorithm dispatch rule research dispatch job resourc work sequenc made base job custom similar shortest process time hundr rule heurist provid review common sequenc rule industri al provid quot survei articl research result sequenc rule summari commonli rule determin job factori rule includ job job total work shop resourc reduc job job job obviou job dispatch base job earliest job arriv job process job result minimum differ workload job enter system assign manufactur assign number dispatch rule conwai perform definit work perform assign rule han recent studi perform complex set rule job job finish resourc rout rout made long advanc process planner common manufactur facil dispatch rout flexibl factori perform potenti obtain common rout rule rout resourc work sit rout resourc finish job rout resourc time energi process job factori control perform offer form manufactur order achiev potenti control factori outlin execut algorithm algorithm factori control put current factori control system implement case case situat agent paradigm open avenu futur research survei organ rest survei organ basic type factori control algorithm type algorithm advantag limit implement algorithm discuss type algorithm implement qualit featur distribut balanc prefer factori control algorithm distribut agent system amount comput requir agent amount comput agent physic prefer agent correspond physic entiti system peopl number entiti number agent physic correspond consist essenti advantag agent local abil peer human oper scalabl prefer algorithm bring system halt agent ad system agent amount comput requir amount comput capabl ad agent ideal situat realiti hope algorithm grow linearli polynomi number agent system survei conclud summar factori control algorithm implement futur research opportun dispatch algorithm dispatch algorithm decid factori resourc avail job algorithm factori wide varieti product advanc plan difficult commonli call factori control control manufactur system manufactur control manufactur plant market inform raw materi finish good market inform sensori inform actuat signal figur control theoret view factori control problem agent repres individu issu agent individu implement control algorithm resourc repres implement control algorithm gener issu warrant discuss paper manufactur research laboratori agent configur individu machin agent make robot modular piec equip robot manufactur technician requir limit effort configur robot applic agent offer excit possibl reduc engin content design build individu piec manufactur equip research discuss survei resolv issu essenti quest develop structur factori remain control issu factori control concern coordin manufactur resourc make desir product algorithm level decid product expend resourc make job job job sequenc control necessarili easili implement paper survei issu factori control high level factori basic type control algorithm dispatch schedul pull algorithm algorithm type differ factori control info scrap materi info materi resourc info transform info storag info verifi control materi transform test inspect command materi storag figur iso gener activ model manufactur system factori control algorithm implement addit research requir implement commonli algorithm invert factori control algorithm agent architectur defin factori control actuat manufactur plant make present past observ state manufactur demand market iso gener activ model manufactur system place control center manufactur system activ figur factori control central nervou system coordin factori give system mean figur show factori control problem form recogniz classic control perspect perform factori control problem split level past control individu manufactur resourc control deliv expect control function control techniqu pid root analysi design implement control algorithm digit analog devic control machin mean apt nc program unit process control chemic industri mean track flow control mean perform motion prescrib procedur control algorithm natur individu manufactur resourc factori control put agent machin made work peopl human level augment perform man machin human focu leverag talent aspect manufactur consist current trend industri local reduc level manag architectur enabl massiv parallel agent wide network agent work solv bring wide distribut comput resourc solv problem agent architectur map internet structur internet featur attract creat agil manufactur system fulli oper system dai instal plug manufactur resourc manag control individu agent function agent architectur integr function island autom softwar agent singl factori spread factori internet form singl virtual advantag emerg local machin oper human level claim blackboard hierarch agent architectur show greater term reduc reduc massiv internet virtual format potenti advantag reduc reduc schedul benefit agent approach provid greatest benefit manufactur chang configur factori ad remov product group manufactur time form virtual consist subset manufactur predict manufactur scenario manufactur manufactur oper significantli grow manufactur distribut oper wide area network particip electron manufactur empow worker provid inform aid make implement distribut intellig factori control factori control algorithm provid basi futur manufactur natur lead question control factori architectur factori control blackboard modifi agent blackboard base applic manufactur includ system system smith boe control system hierarch multipl level agent relationship agent level hierarchi slave master agent highest level hierarchi holon manufactur hierarchi agent call earli research dispatch schedul hierarch architectur paper architectur agent commun fix type agent replic global inform elimin architectur topic paper hierarch architectur differ architectur number import function architectur function agent replic function fix master relationship slave function blackboard architectur consid blackboard global inform storag architectur global broadcast inform inform persist global agent blackboard architectur specif function hierarchi differ fix relationship promin hierarch architectur global inform repositori violat goal elimin global inform manufactur activ research field model biolog system termit hill ant coloni wasp nest bird flock fish school wolf pack inher capabl agent system includ emerg behavior termit build termit hill scale add join process system half termit coloni remain termit continu build termit hill system exhibit call emerg behavior simul termit shown individu termit ident program follow local behavior termit emerg perform global termit hill complex behavior build termit hill emerg combin termit term claim advantag architectur includ reduc reduc cost holon manufactur commun view agent technolog human manufactur machineri work colleagu factori control trend inform flow binari data type basic data type subroutin structur data type function abstract data type oper object tool method agent militari factori messag tabl progress agent technolog ident complimentari agent act number architectur build system architectur type architectur commonli studi system research commun architectur hierarch architectur receiv great deal attent manufactur system research commun architectur specif focu architectur function architectur agent repres function capabl agent function function commun directli main memori main memori network function commun relationship function call function expect specif result commerci softwar function structur analysi design techniqu softwar agent enabl make function agent function architectur develop manufactur problem provid refer blackboard agent area agent share post partial solut problem central blackboard blackboard persist channel past commun persist factori control content introduct introduct agent architectur factori control factori control algorithm survei organ dispatch algorithm review dispatch algorithm dispatch schedul algorithm optim schedul simpl optim schedul mathemat program optim schedul optim schedul simul anneal genet algorithm heurist schedul method schedul determinist simul intellig schedul pull algorithm review pull algorithm pull summari refer introduct agent idea manufactur receiv great deal recent attent attent caught billion intellig manufactur system program decid make manufactur initi test case studi interest unit state nation center manufactur scienc decid initi number program manufactur lead unit state manufactur govern agenc claim manufactur futur manufactur agent technolog foundat technolog implement agil manufactur vision great deal recent interest agent technolog gener numer articl agent technolog societi magazin ieee spectrum commun acm popular press wall street journal time magazin world report major internet comput magazin interest correl recent interest internet comput potenti agent enabl electron commerc inform dissemin retriev internet agent technolog build paradigm tabl agent basic softwar object object system mean commun object unlik softwar specif call agent softwar continu act initi agent system design agent meet meet goal program meet design system call consist group manufactur system sme survei factori control algorithm implement pull albert baker elec comp eng comp sci dept cincinnati ohio paper describ architectur includ architectur review claim advantag describ type factori architectur survei common type factori control dispatch schedul pull algorithm algorithm implement paper describ common factori control algorithm industri implement discuss algorithm popular current research implement requir research system schedul discret distribut artifici agent system introduct paper survei type factori control pull address algorithm put describ abil implement algorithm desir paper provid tutori establish identifi research opportun paper guid implementor past inform past implement found introductori section overview agent technolog definit factori control discuss import pull algorithm factori control introduct conclud overview rest survei 
ML/952 6 target acquisit multipl object behavior coordin paolo maja matar usc robot research laboratori southern california lo ca work fund offic naval research grant abstract paper approach coordin context cooper target acquisit approach base multipl object behavior coordin extend multipl robot mechan distribut command fusion group robot pursu multipl goal multipl robot parallel mechan enabl robot select action benefit benefit group result mobil robot valid group robot successfulli track acquir move target introduct cooper team robot unknown set pose complex control problem requir solut suitabl multitud conflict task object robot action optim respect robot unaccept respect major issu design system control coordin multipl robot problem decid action select robot make benefit group central approach problem attract due inher scale requir commun commun global knowledg environ distribut approach hand appeal due properti scale reliabl overview approach issu cooper mobil robot relat work target observ allianc action select consist inhibit motiv approach similar task includ emploi format control type cooper explicit task paper approach multi robot coordin base multipl object behavior coordin view extens arbitr approach enabl command fusion group robot pursuit satisfact multipl goal multipl robot parallel approach command fusion approach find suitabl multipl conflict robot behavior object successfulli appli approach coordin behavior singl robot extend approach task appli approach task cooper target group robot collect captur move target behavior coordin multipl object behavior coordin base multipl object theori approach make complex situat object sider object defin standard qualiti altern evalu object map assign reflect altern desir multipl object behavior formal object function defin map action interv action fx defin finit set altern action map behavior object function formul coordin multipl behavior multipl object problem seek optim multipl object function object function produc multivalu behavior set altern action action chosen satisfi object function formul multipl object problem argmax subject object function estim behavior control vector xn constrain action space solut equat correspond select action satisfi object behavior solv problem number method literatur solv advantag disadvantag give descript highlight main characterist multipl behavior framework fuzzi behavior object function encod semant fuzzi membership function excel overview applic fuzzi logic robot action space constrain due physic geometr set constraint mechan construct subset feasibl physic action object make make suitabl behavior coordin start describ type solut problem solut optim result maximum object function simultan optim solut equat iff import characterist problem multipl competit improv deterior optim solut exist concept optim modifi includ similar concept problem practic hope solut satisfi object suffici idea encompass definit definit satisf satisf solut extend multipl object problem reduc subset feasibl exce aspir level object point impract search optim solut argu practic effici solut satisf solut correspond behavior maker solut notion optim context multipl object make definit solut multipl object problem iff exist solut solut improv object function simultan deterior object solut nondomin effici solut action select method gener solut notion satisf plai role formul behavior coordin mechan properti techniqu gener po satisf solut problem base definit behavior coordin scheme proce feasibl action base physic hard find set feasibl permiss altern action base object criteria find set action reject reduc significantli set altern satisf action incorpor subject knowledg aspir find set satisf altern prefer action criteria final select prefer action approach systemat narrow set includ increasingli criteria set hard constraint natur task physic limit robot limit exclud set action base object exclud set domin solut simpli effici compar solut incorpor subject prefer structur term set suffici solut criteria subject knowledg includ make final select altern evalu behavior object rank term prefer action relat fuzzi control mechan multipl object behavior coordin fuzzi interchang formul inferenc mechan multipl object formal tool behavior coordin resolv conflict principl manner respect behavior blend fuzzi multipl object approach compat combin straightforward combin approach behavior fuzzi gener multivalu behavior output fuzzi fuzzi context rule gener weight coordin behavior gener satisf solut multipl object make techniqu combin advantag approxim reason behavior specif fuzzi logic qualit formal behavior coordin capabl multipl object make cooper target acquisit appli behavior coordin approach task cooper target acquisit current implement team robot pursu surround target limit robot due practic space equip limit approach easili extend gener case cooper robot group task task robot search surround pursu final captur task high state diagram figur surround search captur pursu lost found lost escap reach approach lost lost found reach dir captur search figur finit state descript target acquisit task group perspect individu perspect success accomplish task effect cooper robot focu cooper surround target cooper enabl effici distribut search target captur focu paper overview cooper search strategi robot task robot task consist search approach captur depict figur order decid direct approach robot requir cooper partner techniqu multipl object behavior coordin enabl collect individu object strategi surround group collect object distribut robot evenli arc target result format shown figur dq target figur strategi group object surround target individu robot translat argmax subject angl describ direct robot target defin window toler robot requir stai equat assum describ complement mean implement equat state robot approach target direct conflict direct partner robot gener desir approach target direct base target rel angl target left robot desir approach left target front desir approach left robot common refer coordin target common refer point suffici target move robot illustr dynam strategi cooper robot case robot approach left assum robot desir approach left robot state desir approach left figur illustr object function robot make process robot approach robot approach left depend multipl object make process chosen robot requir attain aspir level produc satisf solut select solut thick object function exce aspir robot state desir approach target conflict robot desir approach left resolv conflict robot robot desir aspir level figur pursuit direct negoti techniqu multipl object behavior coordin cooper case robot desir approach target direct case experi conflict make situat resolv robot wander approach target direct solut give prioriti case conflict robot highest prioriti author select altern assign weight valu object function robot highest weight win conflict situat hierarchi effici case chosen altern strategi robot decid weight valu base sensori inform robot desir approach target determin rel posit attenu function distanc target object function robot weight proport distanc target impli closer robot target higher weight approach make robot farther room maneuv format robot close target system overview section briefli describ control architectur robot robot platform system implement mobil robot stack pentium processor run linux robot equip soni activ camera connect frame ring ultrason posit compass front sonar soni camera control freedom control angular absolut camera softwar architectur control approach multivalu behavior gener fuzzi express action select perform multipl object behavior coordin lexicograph approach shown advantag combin fuzzi multipl object behavior coordin behavior repertoir describ behavior safeti behavior control veloc base front sonar maintain safe speed behavior avoid behavior control head robot base sonar read maintain safe travel direct left left face behavior servo head robot base input vision face target vision system imag posit target chromat color differ target posit center imag gener error vector behavior attempt minim result robot rotat target center imag front forward left left behavior control pan angl camera angl negoti target approach target robot sensor provid distanc inform size target imag proport behavior trigger target robot field view larg triangular membership function center ang effect combin behavior approach behavior robot target direct camera straight ahead drive robot closer target robot closer turn camera head target align control camera pan angl align robot target desir surround behavior control camera pan angl express desir direct approach target forward left left left output behavior partner robot select direct rotat rotat behavior control angular veloc robot rotat left robot turn direct run circl robot target action select coordin activ behavior combin command blend mechan figur depict behavior hierarchi relationship behavior phase search approach target distinguish visibl target robot field view target search behavior activ approach behavior activ case robot safeti behavior command fusion action select wander safe rotat safeti face avoid search approach align target captur arbitr command fusion action select figur behavior hierarchi action select target acquisit task result section describ valid result trial confin lab environ setup pioneer robot section detail target acquisit manual guid dummi target futur plan autonom robot replac order log posit orient robot experi conduct overhead camera track system capabl simultan track marker posit orient lab show reliabl result success cooper target acquisit trial complet batteri experi progress valid experi target follow search search align target start target end search follow align follow align figur align behavior static target move target conduct simpl experi robot follow target result show robot search find approach direct figur field view robot camera narrow target easi lose robot close robot immedi search direct detect target find figur result valid singl robot follow pursu target group experi robot perform task result robot surround includ cooper make negoti direct robot approach target left target search follow surround follow target surround surround left left figur surround behavior static target move target figur show segment singl trial left target initi static short search robot detect target approach align end stabl format surround target figur target start move robot follow format trail plot tempor occurr event target start move robot immedi hold back path clear proce system mechan explicit coordin result interact robot constraint action select mechan futur work paper extend multipl object behavior coordin approach cooper multi robot task approach easili extend altern command fusion mechan fuzzi conclud work extend command fusion action select mechan coordin contrast behavior arbitr mechan allianc essenti coordin command fusion pursuit multipl goal multipl robot parallel demonstr command fusion produc dynam task context cooper target acquisit result valid approach produc reliabl cooper current implement emploi approach easili scale arbitrari number robot problem problem anticip larg number robot amount inform exchang futur investig strategi robot commun nearest neighbor robot task produc accept result intend investig applic approach distribut cooper search refer ronald arkin robot intellig robot autonom agent seri mit tucker balch ronald arkin format control team ieee transact robot autom uni alex andrew cooper mobil direct autonom multiobject make theori volum miguel schneider maja matar territori task ieee transact robot maja matar issu approach design collect autonom agent robot autonom decemb lynn parker cooper robot observ intellig autom soft special issu robot research oak ridg nation paolo multipl object action select behavior fusion vote phd depart medic informat imag institut electron aalborg fredrik august paolo maja matar approach fuzzi behavior coordin ieee intern symposium comput intellig robot novemb alessandro fuzzi logic autonom robot catalogu technic report universit de av cp novemb herbert simon scienc manag harper 
DB/610 3 scalabl consist protocol distribut servic colleg comput georgia institut technolog ga scalabl inform system silicon graphic mountain ca august abstract common address scalabl requir distribut servic emploi server replic client cach object encapsul servic state perform system depend protocol implement system maintain consist object copi explor scalabl consist protocol requir synchron commun node copi relat object achiev develop approach call local consist lc base protocol provid flexibl effici node control awar updat cach object develop protocol implement strong consist approach demonstr scale tradit invalid base consist protocol system load geograph distribut dimens scale scalabl distribut consist protocol introduct comput local area network larger interconnect growth internetwork highli interact applic span local network exampl applic domain comput support cooper work virtual realiti base distribut shop applic support distribut servic provid low latenc access dynam chang share state problem scalabl consist access import dynam content web concern system support build scalabl distribut servic provid access dynam chang inform explor scalabl servic dimens system load geograph scale system load abil system work support part nsf grant withstand growth total number potenti client geograph scale abil system toler reduc effect larg latenc distribut system common address scalabl requir distribut servic emploi servic replic cach servic state client node assum servic encapsul distribut object scalabl implement servic lead object state distribut multipl node server replic client cach introduc problem consist multipl copi object focu paper develop evalu scalabl consist protocol consist problem import due interact natur assum access share servic state approach explor develop scalabl consist protocol context distribut share memori synchron inform lock exploit reduc cost consist mainten system scale vari coupl applic assum access servic state control synchron oper share virtual node read render scene recent cach copi incur overhead lock oper approach improv scalabl consist protocol provid weak consist share state updat multipl lead conflict updat resolv explor strong consist protocol ensur access object implement relat servic serial node order updat object conflict updat strong consist formul number wai sequenti consist serializ share memori system consist model requir execut oper share object serializ consist condit time oper execut requir updat object node visibl node copi amount time interact time dissemin updat requir provid strong consist model atom memori consist model requir serial oper respect time oper execut protocol implement strong consist permit accept dissemin effect updat oper explor notion strong consist serial object access sequenti consist client node control quickli awar updat cach object flexibl control updat dissemin develop scalabl protocol strong consist develop approach build scalabl consist protocol develop mechan call local consist basic idea lc mechan summar tradit strong consist node updat share object notifi node copi updat node wide updat latenc larg lc base node updat share state inform node node ensur object copi access consist node awar extern updat fetch copi demand accept copi node execut local consist action invalid local object copi overwritten respect newli arriv object advantag approach access latenc lower lc base protocol read write access complet requir commun node copi access object object copi lazili invalid node execut local consist action object number client request server reduc server load consist relat inform batch object copi network load node fetch copi object demand result object copi node introduc cach consist manner push pull style updat dissemin lc base protocol lc base protocol serial object access client control quickli updat visibl currenc cach copi control base resourc avail commun applic evalu lc base consist built object cach framework top system corba distribut object model preliminari perform result support compar lc base strong consist refer sc lc invalid base protocol call sc inv sc lc carri consist action object copi pull demand object access fault sc inv synchron invalid cach copi object object written show version sc lc exploit lc approach read oper complet older object scalabl achiev dimens scale drive system synthet workload captur read write share pattern larg scale demonstr averag time object access sc inv slower sc lc presenc wide distribut sc inv load server client node sc lc sc inv entail messag client node sc lc rest paper organ section motiv lc approach section describ lc base implement strong consist prototyp evalu topic section discuss relat work section conclud paper section assum lock acquir access invalid messag node request lock lc approach system model assum share object store server node client node cach object copi access local object relat consist requir defin relat object relat document object result experi discuss user updat desir user access updat copi copi node copi interest updat object access defin system global view node view captur valu object access node global node view imaginari extern observ updat object instantan global view defin valu object extern observ object observ progress global view node cach subset node view defin valu cach node ideal case node learn updat instantan summar node iew ae current global iew achiev ideal node updat object synchron node copi object wide area latenc make synchron costli node wide distribut node copi updat object interest object work set lc global view updat updat node made advanc view current global view node view lag current global view node view node ensur view belong global view recent intuit show assum system question central access complet send request central server client updat request servic global view node view defin valu return read request current global view node updat share object assum global view chang updat complet global view interest observ result read request result depend request reach process server request global clock exist request made newer factor network delai subsequ read return older valu lc approach base observ defin term node global view node view local consist node iew ae global iew node iew ae global iew node iew ae global iew time global view view figur global local view global iew recent global view global iew node view necessarili subset current global view belong global view currenc node view monoton advanc decoupl node view instantan global view kei provid scalabl consist access share object figur show simpl illustr consist node view node write share object result progress shown global view assum complet write start empti cach read cach notat mean copi object view local view belong global view read ad object copi cach cach copi local view belong recent global view previou view belong write valu updat copi inform object copi view local consist request copi read ad cach action leav cach state shown figur view local consist belong global view maintain copi invalid copi ad cach result belong global view recent view cach consist local invalid overwritten object copi node view advanc consist protocol synchron invalid cach copi write ensur local view node synchron current global view lc base local view lag current global view lag control depend resourc avail applic commun bandwidth client period request updat server push object valu cach node client node receiv valu consist introduc cach perform local consist action lc base protocol provid enhanc flexibl client access consist object copi level currenc cach inform strong consist lc approach develop scalabl protocol implement strong consist protocol ensur access complet node serial order consist order access node protocol focu ensur safeti requir avoid conflict make serial augment addit mechan meet live requir focu mechan protocol ensur oper serial compar protocol invalid protocol sc inv briefli describ protocol featur lc base protocol outlin actual protocol present invalid protocol inv invalid client cach object copi access client write object cach copi mode xs copi experi copi cach experi commun server return copi copi system copi server invalid collect acknowledg respect node return access client refer client cach object copi mode owner attempt read object copi experi commun server client cach object copi server return copi client server downgrad owner copi mode latest copi owner scalabl protocol limit requir commun node object copi write request receiv server lc base protocol present requir commun lc base protocol similar sc inv lc base protocol assum singl writer object time import differ invalid protocol base lc lc base server transfer write access exist reader invalid messag reader object coexist writer read valu consist node view invalid messag node receiv process messag time object cach updat node lc base protocol order access relat object advanc node view consist manner time object copi ad node node perform local consist action ensur store copi share object valid node view consist action requir commun invalid local copi potenti overwritten view client node commun object server object cach updat object read permiss commun server advanc client view client view advanc period commun initi server client period commun ensur lag client view global view bound object access lc base protocol offer flexibl control currenc cach object copi base resourc avail applic aspect consist protocol local consist action implement explor implement action assum object consist desir store singl server node protocol determin object copi remov node cach node view handl distribut server implement address problem relat object store multipl server node protocol present protocol lc approach implement strong consist call protocol protocol base effici sequenti consist protocol develop share memori system protocol similar present requir client commun server write oper protocol share similar approach includ dynam protocol implement sequenti consist memori protocol maintain consist node view receiv inform object copi store node overwritten commun server case singl inform maintain effici server node protocol shown figur write requir commun server server send messag invalid copi system request receiv inform object invalid maintain refer invalid shown method figur client node request ownership object result server determin copi node ad client request return request result request read miss handl sc inv invalid set request client return case write client receiv invalid cach copi object list object copi client invalid client commun server object written simpl execut protocol shown figur client node initi step cach action client invalid local copi object member read definit copi receiv invalid local copi object member read action current owner current owner client cach add figur protocol step node node remark empti empti xs owner ys owner copi downgrad read copi downgrad add ad add remov ad copi invalid read read latest figur protocol simpl downgrad node copi mode step write commun server grant ownership ad note continu read copi contact server step contact server write server add return local invalid includ read copi read copi invalid protocol figur show action advanc client view action execut client server result server send client messag client invalid local copi object includ messag futur access invalid object result commun server advanc client view recent correct correct proof protocol requir demonstr serial object access protocol maintain consist node cach serial ff produc access complet commun ff simpli order central server execut request oper object cach permit client complet access commun server server serial oper requir commun server assum execut oper requir commun server set local access immedi follow serial ff access consist updat local oper client receiv commun write object copi cach access visibl immedi serial construct client read valu older view object updat node detail proof found object lifetim base protocol consist requir relat object store node maintain singl server incur addit commun overhead develop protocol distribut server assum relat object partit server server object remain fix migrat server replic address lc base node view advanc newli cach object exist object copi overwritten node view approach object copi determin exist object copi valid node move newer view denot copi object store call lifetim lifetim object copi time interv copi access violat consist synchron global clock node access object lifetim defin time interv time read written call time time read overwritten gener copi lifetim easili copi object object updat period time associ lifetim object copi written time write object address problem store lifetim object copi associ copi lifetim interv start advanc system learn overwritten object copi cach node lifetim straightforward field includ web page store dynam content lifetim establish valid object test develop determin set cach copi belong consist global view copi belong consist view overlap lifetim written recent greater lifetim copi overlap correct gener belong consist global view lifetim copi local invalid ad node cach lifetim base protocol lc approach node carri consist action local commun node consist preserv cach copi overlap belong common global includ global view incom object copi gener set copi belong consist view pair copi set overlap lifetim defin object lifetim synchron global clock larg scale distribut reason assum exist clock develop consist protocol requir synchron clock logic clock record object lifetim vector clock vector time order write detect conflict updat replic data system bayou coda vector clock vector integ compon updat node updat node defin node object updat denot clock node vector time updat compar node increment increment return maximum vector time vector compon equal compon compon strictli compon vector time compar equal time call concurr kei correct implement strong consist object lifetim read logic clock set rule clock maintain node advanc clock node global view order logic time creat achiev node increment compon logic clock view advanc order view result execut oper develop set clock rule time object copi updat node clock ensur clock time view correctli reflect order view object lifetim similar field alex protocol explor reduc consist relat messag differ logic time defin lifetim ensur consist relat object refer time interv futur client notifi updat cach copi period lifetim refer time interv past object larg number node access assum node access object mode scalabl vector clock issu strictli compon updat node compact logic clock place vector clock valid clock rule order view logic clock clock advanc reflect view tradit vector view result execut oper node order increment node clock execut oper order oper node messag pass vector time includ messag advanc node clock send messag includ time read clock messag receiv time includ messag updat object share oper node order reason oper write read oper order valu written consecut write read oper return order captur order vector clock advanc cach execut ensur time execut properli order order read oper updat overwrit associ denot reflect time read node time read equal object copi section order read return write oper overwrit easili ensur execut overwrit updat execut clock captur node clock action gener object access rule node complet access local cach copi object object copi read complet advanc node remain view access updat object copi node access increment node view advanc result updat rule access result object commun xs server fetch object copi access complet server writer system node cach server send copi server commun writer return writer copi client downgrad access mode receiv copi advanc vt ensur view read access complet order view produc rule object cach mode updat experienc write experienc commun xs server node cach write server request downgrad access current copi advanc vt action ensur write execut order write produc overwritten copi oper read order updat similar write cach increment write complet rule ensur time read vector clock node view order consist order view rule ensur write time object total order rule ensur time access read object greater time write oper execut object maintain rule ensur consecut write time read oper complet order time discuss time assign object copi read valid time node execut oper gener object copi simpli complet cach object mode experienc execut updat xs overwritten ensur vector time view valu gener total order updat clock view oper execut order view updat oper overwrit read multipl node reflect read oper implement server contact client readonli copi return node experi maximum clock valu obtain client invalid base protocol sc inv induc commun client cach readonli copi encount avoid maintain xs server node rule ffl node request inform server local server updat maintain set send request server ffl cach node mode experienc servic xs server commun downgrad copi mode return current server updat receiv clock acquir copi server return experienc write order read copi overwrit read node execut reader coexist writer read complet view time concurr time view result execut perform node oper awar result invalid view result execut concurr view older copi read order read serial local consist perform object copi invalid cach object copi ensur older copi lifetim incom copi access commun server copi overwritten view gener object copi invalid unnecessarili node advanc copi advanc reduc unnecessari invalid object copi updat ffl local cach object copi advanc node clock object copi valid node present view arbitrarili high set object copi written ffl server request node object node updat copi clock return copi reason ffl protocol server return valid copi server updat object copi clock return request client node aggress advanc server node maintain vector clock time client contact server write fault object copi server local clock server updat clock vector time receiv client advanc view reflect oper execut client client request copi object increment wt vt oper wt vt oper figur sampl execut scenario clock send server request object ensur server clock reflect updat oper execut client server make clock advanc valid time object copi send request node protocol consist protocol base object lifetim techniqu shown figur simplifi notat refer copi store object copi relev discuss denot store copi object read write access complet local cach omit descript oper focu client request add object copi cach desir mode handl event shown client side client experi access miss fault object request copi server receiv copi client compar incom copi lifetim lifetim relat object copi cach updat node clock local invalid local invalid ensur node cach consist copi object node clock advanc rule time object copi result commun similar invalid handl differ lifetim base protocol receiv write request client object owner server request owner copi owner statu downgrad copi invalid copi invalid messag client server recent copi system return request client multipl reader coexist singl writer execut scenario shown figur intuit work protocol object initi step node quantiti left individu oper figur object local clock oper execut cach copi downgrad copi updat copi commun server obtain ownership action client node definit object cach local invalid action xs figur lifetim base protocol read copi invalid local copi node advanc node cach copi updat assign rule local copi invalid node continu read older futur read node result cach miss read recent protocol figur includ action advanc client view period similar action ad protocol server send object lifetim record client requir lifetim receiv server higher client advanc copi receiv lifetim higher cach copi local invalid node advanc view request recent copi invalid object correct similar correct lifetim base protocol requir demonstr serial object access protocol maintain consist object copi basic idea produc order access logic time execut clock updat clock base time object ensur logic time oper execut order consist manner correct demonstr argument inspir work scalabl weakli order memori system system assum type synchron label oper ordinari oper synchron oper order ordinari oper result execut obtain strongli consist system assum user program properli access execut acquir lock read slightli stale data accept applic read complet cach copi acquir lock absenc explicit order induc oper clock rule approach result implicit synchron demonstr access serial proof demonstr correct found hybrid protocol object lifetim base protocol conserv requir assum figur write relat object invalid copi execut overwritten write advanc clock execut protocol invalid object copi unnecessarili protocol suffer unnecessari invalid work correctli singl server case outlin hybrid protocol combin protocol call sc lc client experi object contact xs server denot server maintain read valid time object copi invalid set time includ messag receiv client node receiv local cach object includ invalid object manag invalid write object copi contact invalid set inform object maintain correctli object manag consist maintain object lifetim base protocol time receiv copi combin protocol correctli maintain consist cach object copi implement sc lc prototyp system discuss prototyp implement implement object cach system evalu protocol previou section system support base distribut object invok remot method invoc paradigm code state cach local system built top toolkit support rmi base distribut object perform consist action cach object object method execut read object state state updat current version process run client commun server share memori client process share object alloc space cach architectur client node impact object invoc focu paper compar evalu consist protocol omit detail found focu evalu workload absenc trace interact applic deploi consid synthet workload evalu consist protocol exist workload evalu fall domain distribut file initi evalu system base distribut file system workload workload chosen richer share pattern exist workload captur access user cooper geograph distribut softwar develop environ workload matt blaze thesi larg scale distribut call princeton workload document file access approxim user collect week princeton comput scienc depart workload evalu protocol system load aspect scalabl workload larg industri distribut file system support thousand user organ project orient workgroup connect hierarchi local area wide area network spread metropolitan area studi conduct research toronto call workload toronto workload evalu protocol geograph distribut aspect scalabl inform xf trace chose implement synthet workload gener usabl live load trace toronto workload uniqu evalu geograph synthet workload gener paramet number object invoc client make object invoc princeton hour gener time randomli averag time averag time princeton studi approxim forc experi run hour order run experi control time reduc time arbitrarili effect accuraci result discuss section time object access probabl object creat exist workload web server support access dynam content url workload repres singl writer multipl reader access pattern explor consist distribut server object updat multipl node web workload previou read read write write oper user user user user read write tabl file inertia probabl chosen object access minut tabl tempor local file access client read write inertia determin probabl client access file probabl mode access file inertia princeton trace tabl likelihood file written sharpli open read time entropi suggest princeton workload tempor number tabl princeton workload file wide share node good candid replic cach expect chang quantiti workload experi ratio workload gener workload gener event time gener iter number client averag time determin timestamp event gener creation rate determin event object creation object access object server chosen equal probabl case object ratio determin access type tempor local suggest object list event gener client object inertia determin oper object access object entropi forc oper read data structur object inform access access type client number success read list reader writer object number reader writer decid object creat number read share exact ident reader writer determin equal probabl client workload gener faithfulli valu workload paramet previou studi princeton mention time princeton workload reduc time arbitrarili effect result experi attempt trace faith intend workload approach interest reduc averag chose method shorten experi durat client delai access present time match event time track access delai manner number indic amount control workload number delai control time chosen smallest time client delai access ensur access time object node suffici account time server contact writer send consist relat messag perform experi princeton workload averag time time averag time found protocol sensit chang interest compar feel experi reduc access time perform evalu import perform metric studi averag time object total number cach miss server time spent server servic cach miss server request messag server node send servic client request measur server load client product cach miss experienc client time spent server servic cach miss cach miss server request account total number messag system repres network load perform studi report includ action sc lc period requir client commun server advanc view client view advanc client experi invalid protocol sc inv cach copi invalid synchron write object read access sc lc return older valu compar sc inv sc lc aggress downgrad access object copi client updat object client experi commun server node commun result invalid object copi cach replac result commun server replac object access futur result report base sc lc protocol period messag advanc node view overhead messag depend coupl requir node view explor futur work setup total eleven mhz control experi main memori connect atm switch cluster scalabl evalu dimens server load wide area number client cach miss number client figur cach miss time measur commun cost machin locat citi cost estim access time case discuss system paramet affect result experi ffl size data averag size file transfer found toronto studi ffl client defin full cach size reduc number cach miss previou studi cach size half size full cach chosen chose cach half size full cach full cach size case smaller compar full cach file princeton studi file total number object workload smaller due fact object creat workload activ studi support small number object implement lru cach replac polici ffl server cach disk emul object server statist previou studi assum server cach hit ratio ms averag disk access time result present result experi evalu protocol dimens scale system load geograph distribut titl graph workload number server client system figur show cach miss princeton workload case number cach miss lower sc lc sc inv differ impli client request newer object copi number client server request number client figur server request number client averag time number client figur averag time number client server load client number server figur server load client number client server request number client figur cach miss server request sc lc protocol frequent return valu differ number server request sc lc sc inv figur expect sc lc client write request reach server due server send messag current owner object exist sc inv server send invalid client copi cach differ metric show averag time shown figur sc lc incur addit comput overhead client server side local consist check process vector time network case mb atm reduc effect compar higher number server request sc inv case experi time server sc lc case sc inv case time result system wide area commun cost incur result present show sc inv entail messag number client averag time ms case number client figur estim time sc lc sc lc incur process overhead sc inv server client request network desir reduc server load number messag figur plot server load client product cach miss experienc client time spent server servic cach miss sc inv load server client sc lc larger system control differ protocol increasingli number client system geograph figur show cach miss server request sc lc sc inv toronto workload control cluster environ creat trace cach miss server request messag trace simul commun cost determin access time compar advantag similar absolut number lower princeton workload toronto workload demonstr share share workgroup level cach miss server request experi latenc calcul averag time factor latenc measur minimum latenc sun rpc round trip machin locat citi latenc ms ms byte messag workgroup toronto studi locat citi time plot averag time figur messag carri data assum time messag control messag assum time byte messag suggest link bandwidth longer factor latenc case scenario latenc equal disk access time ms plot averag time factor latenc figur calcul averag time factor measur commun cost actual measur provid result lack control environ observ total number messag princeton workload case hold differ averag time dramat sc inv averag time sc lc case scenario link bandwidth longer factor latenc quantiti compar disk seek time demonstr advantag sc lc sc inv dimens geograph distribut discuss result perform studi suggest protocol base local consist approach scale invalid base protocol workload evalu protocol system load dimens scalabl studi invalid protocol sc inv entail messag server load client lc base consist protocol number user differ impact network server load show advantag sc opt sc inv dimens scale workload industri set user work workgroup orient group spread metropolitan area evalu protocol dimens scalabl geograph distribut sc opt reduc network load significantli access time demonstr scalabl lc base experi larger system system control experi quantifi copi access sc lc reduc cach miss due fact copi invalid immedi object updat futur explor combin overhead sc lc messag advanc node view larger system simul explor scalabl larg number node system relat work work relat consist protocol develop system includ distribut file distribut object system distribut share memori system relat system comparison includ due space consist issu context web discuss protocol studi includ base weak invalid poll base protocol strong consist argu strong consist provid read return older valu write result synchron invalid cach copi number cach miss significantli protocol data point experi suggest sc lc permit stale copi small number access read data invalid base protocol develop fundament approach access updat frequenc currenc cach inform control base resourc avail permit multipl write object ensur consist relat object evalu sc lc perform invalid base protocol sc inv applic toler lag node view current global view distribut file system emploi cach emploi perform xf system make invalid base protocol ensur cach consist af similar effect bayou coda system address cach consist mobil environ bayou address level consist session weaker consist level permit scalabl focu scalabl implement strong consist avoid conflict updat object similar vector read write time sc lc bayou make vector time updat control order updat process server bayou consid server replic address partial cach relat object client feel dynam access approach explor client dynam request cach subset object interest coda file system emploi server replic client cach file client work cach copi deal disconnect coda updat serial conflict updat detect resolv system report user transact coda provid serializ delai commit transact valid share goal coda client access cach object commun server serial object protocol conflict updat advanc view protocol client commun server similar reconcili coda thor rover emploi object cach strive provid strong consist conflict access rover lead consist violat user repair effect thor transact execut local cach copi ensur serializ transact valid prior commit interact desir effect local oper visibl user execut observ oper extern world problemat optimist techniqu emploi thor conflict common resolv optimist protocol emploi thor provid perform address consist level object access thor support transact includ sequenc access scalabl share memori system stanford dash softwar implement distribut share memori consist action delai synchron oper complet goal work develop consist protocol enhanc scalabl reduc commun larg number node reli explicit server instanc run client node assumpt server store copi relat object client cach object access synchron lock system address scalabl access object exist system support applic make object implement commun client server cach consist issu address system system implement replic object group commun group commun atom order requir strong consist costli cach result membership chang requir protocol conclud remark explor scalabl consist protocol provid flexibl coupl node copi relat object access serial conflict updat copi develop local consist approach protocol develop requir commun synchron node copi object access protocol base philosophi node control interest local step ensur consist object access node awar access object cach approach node control currenc cach object copi resourc avail applic develop protocol support strong consist work distribut server implement consist group relat object implement prototyp system support cach distribut explor benefit protocol creat control environ accur repres larg scale system develop sophist workload conduct varieti show flexibl offer protocol provid improv scalabl futur work focu detail evalu protocol includ richer workload simul explor local consist base approach implement level consist refer adv hill implement sequenti consist system proc intl conf parallel burn causal definit implement distribut ousterhout measur distribut file system proc symp oper sy blaze cach distribut file system phd synthet workload model distribut system file server acm sigmetr cate alex global file system proc usenix file system schwartz hierarch internet object cach usenix technic jan patterson cooper remot client memori improv file system perform proc patterson quantit analysi cach polici scalabl network file system proc acm lori notion consist predic lock system commun timestamp system preserv partial order australian comput scienc hennessi memori consist event order scalabl multiprocessor sandhu case studi file system workload distribut environ toronto technic sandhu case studi file system workload distribut environ proc acm grai effici mechan distribut file cach consist proc acm symposium oper system principl page implement replic file system usenix june seltzer web cach consist usenix technic januari mauric wing correct condit concurr object acm transact program juli howard scale perform distribut file system acm tran comput de kaashoek toolkit mobil inform access proc lazi consist softwar distribut share memori proc intern symposium comput satyanarayanan disconnect oper coda file system acm symp oper sy system support scalabl servic phd georgia inst object cach corba system usenix system lamport make multiprocessor comput correctli execut program ieee tran sep build reliabl distribut system corba proc usenix confer technolog wood dynam reduc coher overhead multiprocessor annual intern symposium comput safe effici share persist object thor acm liu cao maintain strong consist web proc intern confer distribut misra axiom memori access asynchron hardwar system acm transact program languag januari zhou sequenti consist distribut system proc intl workshop theori practic dist system time global state distribut system intern workshop parallel distribut mogul hint cach web acm sigop highli scalabl system proc schiper causal consist sequenti consist share memori system irisa welch session weakli consist replic data ieee symp parallel dist inform manag updat conflict weakli connect replic storag system acm plausibl constant size logic clock distribut system workshop distribut algorithm octob wang anderson wide area mass storag file system workshop workstat oper oct invalid larg scale network object cach master thesi 
DB/2489 3 index move point agarw lar arg jeff erickson abstract index scheme store set point move linear queri form answer rectangl real tq report point lie insid time tq present index structur constant disk block answer queri report point lie insid time interv point insert trajectori point present gener approach improv queri time queri arriv chronolog index evolv time obtain tradeoff queri time number time index updat point move describ index scheme number requir answer queri depend monoton differ tq current time develop effici index scheme answer approxim queri move point center geometr depart comput duke box nc support part nation scienc foundat grant armi research offic muri grant sloan nation scienc foundat award match fund xerox grant scienc foundat center geometr depart comput duke box nc larg support part nation scienc foundat grant depart comput il jeff support part nation scienc foundat grant armi research offic muri grant sloan fellowship introduct effici index scheme support rang search variant central larg system relat system rang search commonli oper problem crucial support languag constraint queri languag class hierarchi spatial geograph inform system rang search plai pivot larg number extern data structur answer queri develop refer store process continu move data wide rang includ digit mobil commun system exist assum data constant explicitli suitabl queri continu move object continu updat queri output obsolet approach repres posit move object function posit explicit chang system updat function veloc object recent work extend capabl exist system handl focu develop effici index scheme store set move rang queri locat futur answer quickli queri point lie insid queri rectangl minut problem statement fp pn set move point time denot posit time fp pn assum point move straight line fix trajectori point chang time assum object updat valu system notifi valu chang term current time interest answer queri rectangl time report point lie insid time report rectangl time valu report point lie insid time report queri point oe time report nearest neighbor oe time point min main interest minim number disk access answer rang problem standard extern memori model model assum disk access transmit contigu block unit data singl oper effici data structur measur term amount disk space unit disk number requir answer rang queri interest solut output queri bound express term number point term number point report queri note minimum number disk block store point report output point refer bound introduc notat previou result detail summari earli work tempor found survei paper salzberg tsotra earli work concentr data activ develop data model queri languag support continu move object sistla wolfson develop tempor queri languag call futur tempor logic support proxim queri move object sistla al data model call move object move object refin model extend incorpor import includ uncertainti motion commun cost model found number practic method access search move object worst queri output size kollio al effici index base partit tree store set point move real line disk answer queri type arbitrarili small constant insert delet point present scheme disk block answer queri type assum speed point practic data structur point move worst case answer queri kollio al practic data structur answer queri move point real bound queri time prove salteni al base index scheme answer queri type comput geometri earli work move point bound number geometr structur convex delaunai point move semin al introduc notion kinet data structur work lead interest result relat move includ result kinet space partit tree cell main idea kinet framework point move configur kinet updat perform data structur event occur point move combinatori structur data structur chang discret time call updat data structur continu contrast fastest move object determin updat time step entir data kinet data structur base natur interpret term underli structur work kinet data structur intern memori comput model result paper main result index move point plane essenti approach approach time axi solv problem space base work kinet data point index scheme evolv time modifi kinet event point equal data structur necessarili updat structur develop index scheme provabl perform bound answer rang queri point move section describ gener concept comput geometri section present index scheme base partit tree disk block answer queri type arbitrarili small constant point insert delet index chang trajectori point index answer queri type point report index scheme handl form uncertainti veloc affect asymptot perform partit tree scheme cost queri high section show index evolv answer queri provid queri arriv chronolog order achiev kinet framework extern rang tree structur log log disk block kinet rang tree slightli develop set al show combin kinet rang tree partit tree obtain tradeoff queri time number event kinet index updat paramet nb answer queri provid trajectori point event queri futur critic queri futur index scheme fast time queri time queri section index scheme log log disk answer queri type number requir monoton function queri time exce point trajectori uniformli queri number kinet event time interv section describ index scheme answer queri paramet ffi construct index base partit disk block queri point oe time return nearest neighbor point insert delet geometr preliminari order state result concept result comput geometri dualiti dualiti popular power techniqu geometr map point line dualiti dual point line dual line point oe denot dual object set object foe oe essenti properti dualiti point line dual line dual point figur dual strip oe vertic line segment oe point li oe dual line intersect oe dual primal figur dualiti transform dimens dual point strip line vertic line segment extern partit tree partit tree commonli intern memori data structur geometr rang search index scheme base partit origin extend extern memori set briefli summar emphasi slightli improv perform compar set point partit set pair disjoint subset triangl point note point lie belong figur size denot number pair partit balanc subset point cross number partit maximum number triangl cross singl line figur balanc partit size lemma state fast partit construct bound mention smaller lack omit proof lemma set point posit integ construct balanc partit size expect cross number construct partit tree node partit tree subset point simplex root js dn node construct root leaf store point singl block intern node constant comput balanc partit rv rv small cross number construct partit tree subset store vertic pointer root ith child block store node chosen leaf node point total number node tree total size partit tree appli lemma build entir partit tree total construct time log expect describ complic determinist algorithm construct balanc partit cross number algorithm requir simpler construct lead asymptot queri bound answer queri find point insid queri strip oe order visit top fashion node report point lie insid oe test triangl li complet ignor li complet insid report point ith oe cross visit ith child note point oe report shown queri arbitrarili small constant make extern partit tree dynam partial rebuild techniqu overmar node store number point store insert point root increment add subset find triangl partit insert triangl smallest delet process similar order queri time static rebuild part partit tree updat node unbalanc child parent leaf unbalanc special constant ensur node tree balanc insert delet rebuild root unbalanc node closest root rebuild root node log expect increment time rebuild amort cost modifi expect insert delet amort cost insert delet bound index scheme kollio al obtain result rang improv updat perform theorem set linearli move point paramet preprocess index size block queri answer data structur construct log expect point insert delet amort cost expect index describ index scheme answer queri point index scheme easili figur rectangl queri move point strip queri static line intersect rectangl point lie insid strip gener higher breviti describ case set point point trace line queri line intersect rectangl plane parallel map line point partit tree answer queri partit tree observ line intersect project intersect appli dualiti transform move point induc static point dual dual subset denot point dual dual queri rectangl induc queri strip oe oe result queri set point oe oe figur construct data structur build primari partit tree point node attach secondari partit tree point attach secondari tree node depth multipl ff log ff constant determin size singl secondari tree total size secondari tree construct log expect insert delet point expect algorithm answer queri attach secondari partit tree node result index requir log block asymptot queri bound structur partit tree queri strip oe oe search primari partit tree point oe find triangl li complet insid oe perform complet search search level secondari structur node secondari tree report point oe denot number requir answer queri secondari partit tree exclud report point insid queri rang shown obei recurr solut arbitrari small denot number answer queri entir data exclud requir report output descend node visit perform ff log ff log secondari point obei solut recurr make arbitrarili close ff appropri theorem set linearli move point paramet preprocess index size block queri answer index construct log expect point insert delet amort cost expect remark gener construct number dimens build level partit tree preprocess queri updat time multilevel partit tree constant factor addit dimens point distribut uniformli rectangular simplifi construct regular squar grid partit detail full paper extern partit tree handl form uncertainti veloc locat point queri perform queri type point ly rectangl time interv report line intersect box report line separ report line intersect top facet left facet front facet separ copi earlier index theorem set linearli move point paramet preprocess index size block queri answer index construct log expect point insert delet amort cost expect chronolog queri discuss improv queri bound present previou section data structur chang queri current time approach extend handl queri futur time long arriv chronolog order kinet rang tree base data structur suboptim queri time develop al extern rang tree develop present rang tree primal set give overview extern rang tree section discuss kinet section main idea store snapshot move point point perform kinet updat data structur section discuss tradeoff queri cost cost updat data structur extern rang tree extern rang tree present structur primari structur tree point similar log tree height log log point store secondari structur node path root leaf store secondari structur linear structur log log block total secondari structur extern version prioriti search answer rang report point ly rectangl form basic structur extern prioriti search tree node auxiliari structur point linear support queri updat breviti discuss structur full descript extern rang tree prove lemma set point store index log log disk rang queri answer point insert delet amort cost set point order order structur answer rang queri consist block store constant number catalog block block catalog block store block point condit block construct creat block ib lower endpoint point ib ib sweep horizont line upward block track number point ly sweep line sweep line reach point consecut block fewer point ly line block creat lower endpoint interv point lie line upper endpoint set longer track sweep sweep continu manner line reach point block creat shown structur construct answer queri load catalog block main memori identifi block includ block sweep line intersect interv load block main memori time report relev point shown queri figur figur answer queri dot line initi block vertic solid block activ sweep line activ block left activ block load answer block consecut pair block contribut point output structur easili made dynam global rebuild updat simpli log extra block entir structur rebuilt scratch amort cost insert delet spend extra examin ensur queri result consist record updat lemma set point store index size rang queri answer index construct point insert delet amort cost kinet rang tree discuss extern rang tree modifi work move point rang queri current time answer quickli explain modif structur answer queri point detail full paper recal endpoint block defin point call signatur case move defin time interv block chang continu time point continu satisfi condit time content block chang discret instant time valu equal hold call kinet event handl kinet event handl dynam global rebuild maintain extra updat block point leav block delet add full reconstruct entir data structur scratch amort cost event point comput time ij leav store time valu prioriti call event track event min ij updat data structur updat prioriti queue point leav enter requir comput failur time ij store event queue event queue updat event modifi scheme chang trajectori point omit detail answer queri time simpli load catalog block calcul block time load block intersect interv time intersect time report point lie queri rectangl queri procedur report point lemma set point move fix veloc store index size current queri answer amort cost event trajectori point chang total event similar evolv primari secondari structur extern rang tree event defin time instanc point equal event calcul effici store global prioriti queue event comput data structur updat event approach similar techniqu al kinet rang tree detail full paper theorem set linearli move point preprocess kinet rang tree size log log block current queri answer amort cost kinet event trajectori point chang total number event remark kinet rang tree work bound point move complex provid quickli determin move point cross horizont vertic line tradeoff perform effici spend total log time evolv kinet rang tree combin kinet rang tree extern partit tree obtain tradeoff cost answer queri total number event construct similar gener techniqu establish tradeoff data structur size queri time describ tradeoff structur move briefli describ gener dimens detail defer full paper paramet structur move point consist top level extern partit log cb fanout tree tree leav leaf store subset point kinet rang tree figur level kinet rang tree partit tree figur schemat pictur tradeoff structur kinet rang tree undergo kinet total number kinet event entir structur comput number requir answer modifi earlier recurr ad base case solut modifi recurr note set constant make arbitrarili small queri time combin structur theorem set linearli move point paramet bn preprocess index size block current queri answer trajectori point amort cost event gener result move point partit tree section modifi primari structur replac level secondari kinet rang tree node secondari structur modifi replac level kinet rang depth omit analysi extend theorem set linearli move point paramet bn preprocess index size log log block current queri answer trajectori point amort cost event index partit tree advantag answer queri cost answer queri high kinet rang tree answer queri handl queri chronolog order present index scheme combin advantag answer queri number answer queri small queri time close current time describ answer queri futur past extend approach arbitrari queri time queri observ previou combinatori structur kinet rang tree chang point equal updat data structur sequenc event time event version event notat defin valid time interv log current time maintain version maintain past version futur version note futur version tent built anticip futur base current trajectori point trajectori point futur version structur chang store store idea persist data structur call data structur data structur recal level primari secondari structur variant store version adapt standard techniqu lack space discuss modif tailor techniqu focu auxiliari structur store node secondari answer rang queri set point recal maintain extra updat block record kinet kinet event reconstruct version discard maintain multipl version modifi updat queri procedur point enter leav record time event updat block addit ident point discard declar inact kinet event maintain sequenc version activ inact version store death time time notat conveni set activ version maintain version disk block report point lie rectangl time find version activ time maintain multipl version primari secondari structur kinet rang found simpli queri section report point updat block insert report point delet total rj store death time version auxiliari structur global prioriti queue ith kinet event time delet version auxiliari structur death time lie interv maintain log version event perform updat oper turn impli perform log updat oper auxiliari structur total size data structur log log spend amort event updat chang trajectori point updat version auxiliari structur store store auxiliari log updat data structur worst case practic smaller omit theorem set linearli move point preprocess index size log log queri answer provid log event current time amort cost event point store place updat trajectori log significantli simpler data structur develop set point case directli appli idea obtain result detail full paper theorem set linearli move point preprocess index size queri answer provid event current time amort cost event point store place updat trajectori log answer queri combin partit tree interim persist index obtain index queri cost function jt combin similar multilevel partit tree discuss section describ index scheme similar approach work set point move fix veloc convert set line denot set point dual line construct partit tree section node subset point set input point node depth multipl ff log ff small construct interim persist index maintain time interv valid total size result structur disk block event defin time instanc point store secondari structur collid event data structur updat point lie interv answer start section node secondari structur report point requir js visit children section maximum number answer independ worst secondari structur close current queri procedur visit level partit tree switch secondari kinet structur difficult bound queri time assum distribut trajectori event occur short period time assum point trajectori uniformli distribut point uniformli distribut unit simplifi partit tree prove theorem set linearli move point preprocess index size block cost queri time monoton function jt point trajectori uniformli queri expect number event prove similar slightli bound size data structur updat time detail full version paper theorem set linearli move point preprocess index size log log block cost queri time monoton function jt point trajectori uniformli queri expect number event neighbor search section briefli sketch index scheme answer queri main idea approxim euclidean metric metric unit ball regular polygon edg polygon defin distanc function dp dp ag easili verifi function metric polygon central symmetr origin ffi regular circumscrib radiu center origin vertex figur dp easi calcul show dp figur figur regular center origin differ euclidean metric wedg qa nearest neighbor distanc function triangl connect vertex origin dp min find nearest neighbor dp suffic comput nearest neighbor queri point distanc function index scheme approxim nearest neighbor queri consist separ data structur loss point wedg form rai eman direct ov ov figur nearest neighbor point oe distanc function point nearest oe direct normal figur point dot product vector normal easi prove nearest point point minim problem hand preprocess set move point plane answer queri point oe time comput point minim solv modif partit tree section construct partit tree project line ov node construct secondari partit tree project subset line ov attach tertiari structur node node secondari partit tree depth tree multipl ff log small constant ff subset point store defin min lower envelop function graph convex chain comput log store chain comput insert delet point updat graph amort cost log total size data structur block queri point oe time answer queri comput node vm secondari tree queri algorithm section comput final return point attain min omit obtain lemma set linearli move point paramet preprocess index size block queri answer index construct log expect point insert delet amort cost log expect data structur queri consist separ copi triangl theorem set linearli move point paramet ffi preprocess index size block queri answer index construct log expect point insert delet amort cost log expect approach extend higher dimens dimens requir level partit order maintain approxim number facet polyhedron approxim unit sphere comput approxim nearest neighbor move point index size queri point detail full paper paper present effici scheme index move point plane queri type answer effici tradeoff queri time time spent updat index scheme point move present effici index scheme answer queri conclud mention open develop effici index scheme point move nonlinear trajectori develop effici index scheme answer exact queri index scheme section extend theorem hold distribut point trajectori refer effici search linear proc annu acm sympo principl agarw geometr rang search advanc discret comput geometri pp kinet binari space partit intersect segment disjoint proc annu sympo discret track index optim rang search proc annu acm sympo principl arg optim dynam interv manag extern proc annu ieee sympo found comput data structur mobil proc annu sympo discret proxim problem move proc annu acm sympo comput asymptot optim vldb battl paradigm time proc intl sympo command control research chomicki geometr framework spatiotempor proc intl workshop time dd make data structur journal comput system erwig develop spatiotempor queri dexa abstract discret model spatiotempor data acm gi access comput comput proc annu ieee sympo found comput index data model constraint journal comput system index mobil proc annu acm sympo principl nearest neighbor queri mobil spatiotempor koubaraki tractabl queri answer indefinit constraint basic result applic queri spatiotempor effici partit discret comput nievergelt spatial data concept design algorithm foundat gi van lectur note comput scienc mh design dynam data lnc mh overmar van mainten configur comput syst sci index trajectori move point chorochrono tech ramaswami oodb index class academ ramaswami path techniqu optim extern proc annu acm sympo principl index posit continu move proc sigmod int conf manag salzberg comparison access method tempor technic sequenc geometr cambridg sistla tempor condit integr constraint activ proc sigmod int conf manag model queri move proc int conf data dynam attribut index comput journal pj rm effici access ieee tran knowledg data onlin data structur extern proc annual intern colloquium lnc cost model posit move proc intl conf data track mobil unit real proc int conf updat queri track mobil distribut parallel move object proc sigmod int conf manag move object issu proc sympo spatial 
DB/1186 3 annal mathemat artif comput extend abduct transact program katsumi inou sakama depart electr electron engin kobe kobe japan depart comput commun scienc japan explain observ nonmonoton background remov addit observ explain order formal extend abduct introduc inou sakama gener tradit abduct comput neg explan remov unexplain neg observ comput mechan extend abduct background theori written normal logic introduc transact program comput extend abduct transact program set product rule declar addit delet abduct abduct explan comput fixpoint transact program model gener procedur correct procedur shown class acycl cover abduct logic program context deduct transact program declar specif updat nonmonoton updat introduct motiv background abduct infer recent recogn import form reason ai logic program tradit formal abduct defin explan observ set background logic entail observ background knowledg base observ tradit framework abduct comput inou sakama comput extend abduct set satisfi consist background theori theori monoton framework abduct background knowledg base nonmonoton knowledg base written nonmonoton logic observ tweeti good reason assum remov fact knowledg base explain observ tradit abduct situat abduct introduc knowledg hypothesi remov knowledg base cope inou sakama introduc notion explan background knowledg base observ set call neg explan consist explan satisfi call posit explan notic fly entail knowledg base block deriv situat notion unexplain observ background knowledg base observ set call set call neg extens tradit abduct call extend abduct paper extend abduct nonmonoton theori background knowledg base justifi inou sakama comput extend abduct reason introduct neg explan natur nonmonoton theori addit formula axiom delet previou theorem nonmonoton delet ad formula delet theorem case delet formula introduc formula nonmonoton reason posit neg explan plai complementari role account observ nonmonoton theori introduct account neg observ monoton theori tradit abduct concern explain posit observ neg observ perceiv situat induct system posit neg exampl construct hypothet theori notion plai dual role explan extend abduct enhanc reason abil tradit essenti theori nonmonoton theori chang abduct theori applic extend abduct includ view updat deduct contradict remov diagnosi repair diagnosi construct diagnost hypothesi identifi minim set fault compon set system compon comp comp consist system descript repres compon faulti repair compon observ diagnosi posit neg explan time al appli extend abduct formal system repair problem context model success techniqu verif concurr program protocol extend abduct extens tradit abduct recogn proof procedur extend abduct anticip comput method extend previou procedur tradit abduct natur outlin studi comput aspect extend abduct knowledg base written normal logic program introduc program transform produc transact program inou sakama comput extend abduct logic program complet program transact program kind disjunct logic declar specifi relationship liter express request addit delet abduct abduct infer comput trace rule program direct complet observ abduct comput clark complet tradit abduct transact program natur extens previou comput posit explan neg explan show procedur comput abduct explan transact program abduct explan observ program comput fixpoint transact program model gener procedur procedur consid extens abduct procedur inou sakama posit observ explain ad transform abduct logic program disjunct program paper notion mark liter introduc paper make liter rewritten procedur correct procedur shown class acycl program ground case class cover acycl program case special correct program present applic transact program updat context deduct transact program declar specif updat special transact program program marek truszczynski extend version rest paper organ section outlin extend abduct framework section introduc transform abduct logic program transact present fixpoint comput extend abduct section address comparison relat section conclud paper abduct framework abduct framework consid paper defin normal logic program finit set rule atom denot negat failur side rule side bodi rule empti call integr constraint inou sakama comput extend abduct empti head integr constraint repres special atom rule empti bodi call fact fact identifi atom program ground variabl nlp acycl level map ground atom natur ground rule level higher level horn rule negat hold rule form definit horn program integr constraint stabl model semant nlp nlp set ground ground horn nlp defin rule ground rule form note possibl infinit set ground rule uniqu minim model model ident call stabl model stabl model consist satisfi integr constraint nlp consist consist stabl model consist nlp atom entail includ stabl model note acycl program uniqu stabl model coincid model abduct logic program defin pair nlp set atom call abduc instanc element call abduc written loss assum rule abduc head fact alp acycl acycl definit ground atom pair call explan posit observ wrt set ground instanc rule abduc bodi made introduc rule abduc translat abduc similar transform appli fact emploi fact abduc inou sakama comput extend abduct posit neg observ simpli call observ call posit call neg call mix minim impli definit restrict version gener framework extend abduct autoepistem theori nlp alp definit result paper directli applic abduct logic program classic negat section number explan observ explan select minim criterion syntact form occam minim explan interest nonmonoton theori chang consid minim minim reduct background theori minim augment newli ad minim accomplish minim chang theori minim paper note hold minim introductori program section alp minim explan minim comput extend abduct provid method comput extend abduct section introduc notion transact declar chang observ hold section describ procedur comput extend abduct rewrit mark liter rule transact program section program assum result extend program section transact program introduc notion transact obtain transform abduct logic program transform motiv inou sakama comput extend abduct rule knowledg base observ hold atom formula call liter transact program consist product liter sequenc liter vertic line read mean sequenc mean conjunct liter product form side call side call consequ product note conjunct product form product written multipl product disjunct liter consequ common product form stand product set product disjunct logic program negat proposit atom definit ground alp transact program defin atom rule rule transform transact product equal inequ consequ definit inou sakama comput extend abduct defin includ case rule head bodi assumpt transform defin empti atom head rule atom introduc product gener step specifi abduct specif explain explain unexplain unexplain mean introduct remov correspond introduc empti string logic true ident set mean remov empti bodi imposs note integr remov mean satisfact constraint abduc requir add transact request remov addit product form mean remov impli requir rule product form mean addit imposs inou sakama comput extend abduct fixpoint comput comput extend fixpoint comput transact program fixpoint oper defin similar inou sakama disjunct logic program oper li notion mark view product rewrit rule liter rewritten mark recogn requir translat request atom formula call liter symbol call mark liter unmark transact program set ground languag herbrand base transact subset herbrand interpret transact specifi atom ad delet transact inou sakama comput extend abduct coher atom includ pair transact satisfi hold conjunct write transact satisfi product impli defin map set transact mark filter introduc transact denot liter mark mean liter remov includ mark unmark form conjunct liter lm denot set conjunct lm definit ground transact set transact map defin coher map defin product set product intuit read definit transact satisfi product reject product satisfi satisfi satisfi consequ expand ad singl disjunct consequ satisfi inou sakama comput extend abduct disjunct expand liter filter ad case simpli mark ad product ordin power defin successor ordin limit ordin definit mean limit ordin closur retain transact persist comput transact ordin smaller includ closur definit comput minim model disjunct logic program result fixpoint transact program fixpoint closur note fixpoint comput construct minim model manner product view disjunct model gener procedur similar comput mechan perform expand herbrand interpret program closur comput procedur stop finit step ground transact condit satisfi readi comput extend abduct fixpoint transact program call abduc liter program rule variabl head occurr posit atom bodi inou sakama comput extend abduct transact set unmark abduc liter denot set transact min denot transact program observ mean integr constraint satisfi integr constraint ad write definit acycl alp ground observ abduct step induct defin explain explain unexplain ground rule head unexplain explain ground explain unexplain unexplain ground rule atom unexplain explain max note definit provabl inou sakama comput extend abduct posit semant acycl nlp model model stabl model coincid acycl nlp hold acycl nlp stabl model provabl lemma acycl ground set ground instanc consist wrt minim wrt proof proof obviou induct abduct step minim explan wrt discuss provabl explain explain minim explan satisfi explain explain minim explan satisfi explain proof minim shown similar theorem ground acycl ground observ minim explan wrt transact min minim wrt transact min proof prove transact min min prove induct abduct step result follow lemma explain min min inou sakama comput extend abduct unexplain rule head min hold min result hold step explain minim explan ground rule explain unexplain induct minim explan min minim min product equal impli coher minim impli minim min unexplain minim ground rule form atom unexplain explain induct minim min minim explan min product inou sakama comput extend abduct disjunct liter product rewritten product conjunct liter form collect disjunct disjunct min min impli coher fixpoint minim impli minim min show put min min show explain proof backward corollari ground observ explan ground acycl alp figur comput explan inou sakama comput extend abduct min figur minim explan program theorem hold gener program explan min impli absenc explan acycl condit theorem program explan includ min correspond explan section show correct procedur comput neg assur class horn program theorem program alp result section directli appli abduct logic program classic negat extend logic program neg liter posit liter rule form answer set semant emploi semant program definit unchang addit requir inou sakama comput extend abduct definit coher transact coher transact includ complimentari pair liter case extend result previou subsect program alp cover nlp observ ground call cover alp rule variabl bodi head note integr constraint cover program ground transact comput ground assum predic atom abduc call predic abduc predic case gener case theori predic complet logic extens technic scope paper presenc variabl clark complet produc transact program recal clark complet defin rule written form variabl variabl origin rule note cover consid rule predic written form equal origin bodi definit alp transact program defin definit base clark complet cover program call constrain program inou sakama comput extend abduct definit cover alp transact program defin predic rule rule transform defin definit includ abduc predic rule rule form assumpt rule transform predic definit rule predic introduc essenti ground case transact program equal inequ cover program ground regard equal inequ liter evalu comparison true term instanti liter ident product evalu check recal rule abduc head fact section inou sakama comput extend abduct instanti unifi equal product occurr consequ product substitut term instanti unifi consequ occurr disjunct consequ substitut term instanti note nlp definit reduc definit map definit slightli extend variabl product instanti substitut transact mark perform instanti liter product obtain substitut match full unif nlp cover observ ground start ground observ fixpoint oper repeatedli appli long transact chang transact construct fixpoint comput ground liter model gener procedur introduc section case constrain fixpoint closur defin manner reach fixpoint result theorem directli extend comput acycl program theorem cover acycl ground observ minim explan wrt transact min minim wrt transact min proof proof theorem lift case includ inou sakama comput extend abduct posit observ min minim explan neg observ includ min minim alp section posit observ min minim explan neg observ min correspond minim discuss abduct procedur idea comput abduct transact program close comput abduct clark complet inou sakama comput extend abduct essenti previou work point procedur intend comput neg mix extend design comput posit explan product newli introduc transform clark complet procedur enabl comput neg observ treat atom complet formula gener transact program declar specifi abduct program gener work procedur call abduc abstract form naiv implement result procedur predic rewritten redund rewrit predic maxim avoid comput mark sound complet hierarch nlp acycl nlp strictli includ hierarch nlp classic proposit theori consid case fulli consid form definit rewrit rule equal fixpoint semant abduct logic program introduc comput tradit abduct comput extend transform alp disjunct logic program paper shown paper similar fixpoint oper comput extend abduct transact program kind disjunct program fixpoint comput realiz model gener procedur provid function rewrit mark realiz procedur abduct procedur abduct nlp abduct procedur develop comput posit explan comput posit comput rule introduc atom obtain credul explan neg mix directli comput abduct procedur damasio pereira abduct framework semant abduct perform system remov contradict program chang abduc liter chang true correspond remov abduc neg explan credul includ stabl model definit explan consid paper skeptic acycl program stabl credul skeptic explan coincid acycl alp inou sakama comput extend abduct mix unknown abduc due roughli correspond situat true assign abduc stabl model assign stabl model abduct framework comput procedur produc transact program procedur comput explan minim abduct explan gener fixpoint oper definit involv comput explan minim expand transact parallel model gener procedur design necessarili solut time expand transact definit breadthfirst manner transact program simul queri process acycl nlp queri posit observ fixpoint comput correspond queri process queri succe refut fixpoint empti set updat abduct updat deduct updat request fact translat updat fact updat call view updat view updat problem present deduct defin consist nlp set ground fact languag call fact insert ground fact delet accomplish updat satisfi satisfi condit condit updat accomplish insert condit present updat minim fact store relat updat request correspond posit neg observ fact correspond abduc view updat extend abduct inou sakama comput extend abduct lemma ground atom updat accomplish minim result obtain theorem theorem cover acycl ground atom insert achiev updat transact min delet achiev updat transact min comput updat fixpoint comput transact program kaka mancarella updat abduct procedur comput view updat procedur explicitli exist fact reason rule fact provid inform simplifi transact remov condit satisfi transact inform edb repres product fact kaka mancarella approach defin presenc integr make consist check candid solut candid inconsist integr constraint prune solv resolut decker satisfi updat request maintain integr integr constraint transact program integr treat uniform manner comput transact program abduct translat disjunct program updat realiz comput updat procedur procedur base tradit abduct produc transact program rossi naqvi clark complet comput updat horn updat transform formula obtain part result formula repres updat origin formula comput procedur lloyd provid procedur perform updat nlp comput updat procedur correct program requir check updat drawn perform updat request procedur produc transact inou sakama comput extend abduct concern satisfi updat gener topic find explan observ consol al discuss similar view updat appli abduc procedur view updat produc transact program inform edb incorpor comput view argu incorpor illustr abduct view updat extend abduct regard remov abduc part abduct grant al translat view updat set disjunct fact base updat achiev disjunct fact fernandez al realiz updat construct minim model satisfi updat request treat ground program possibl disjunct transact program produc transact program specifi updat disjunct introduc disjunct transact program view updat languag declar specifi dynam program transform automat gener updat specif origin program gener transact program kind disjunct updat natur express disjunct product transact program provid applic disjunct logic program note updat languag exist disjunct languag view delet definit aravindan baumgartn transform ground definit program disjunct program comput view delet definit transact program similar realiz view delet comput correct fixpoint comput horn program theorem ground horn ground neg observ minim neg wrt transact min proof lemma delet accomplish minim wrt negat failur neg definit consist product inou sakama comput extend abduct updat request delet concern transact program transact unmark liter case abduc remov abduc definit includ fixpoint product ls remov accomplish delet ls remov translat request delet liter case case abduc case ls remov successfulli accomplish analyz case chain eventu termin abduc ident number rule finit util definit abduct step definit unexplain theorem appli program acycl entail unexplain remov neg minim oper min fixpoint proof direct shown base similar argument alp inou sakama comput extend abduct figur comput explan shown delet illustr figur minim minim neg point gener contract algorithm includ view delet definit program requir comput minim posit explan sentenc contract wrt background knowledg abduc hit set correspond neg method directli comput hit set comput posit explan explicitli contrast transact program appli program negat comput posit extend abduct program marek truszczynski defin languag knowledg base mean program similar stabl model semant logic program defin knowledg base set proposit atom program consist rule inou sakama comput extend abduct proposit atom rule intuit mean constraint belong belong belong belong rule regard proposit logic minim model pair call chang initi dr kb program obtain delet rule form dr dr delet bodi remain rule chang dr call dr shown perform minim chang justifi rule direct arrow rule product similar syntax proposit transact program disjunct program rule liter bodi framework kb rule notion abduc chang repres chang atom includ atom languag insert delet notion mark repres liter transact translat atom inform comput program program atom subject chang view updat consid util transact program program section typic applic program updat recal nlp divid part rule bodi part fact initi kb set stabl model includ fact edb program construct updat request transform transact program kb dr comput transact program notic dr satisfi idb minim edb inou sakama comput extend abduct rule transform set initi kb stabl model insert put view dr obtain chang dr dr view transact includ ignor mark obtain coincid dr case note ignor smaller explan framework treat fact extend program view transform method gener program origin program deduct program construct idb applic program background knowledg base includ causal inou sakama comput extend abduct gener law author method gener program discuss decid rule obtain paper present method comput extend abduct abduct logic program transform transact program declar specifi abduct abduct explan comput fixpoint transact sound complet acycl cover alp context transact program declar specif enabl comput view updat construct manner transact program kind disjunct program fixpoint comput realiz model gener procedur paper assum acycl cover nlp program variabl futur work includ relax condit exploit applic extend abduct refer intern confer lectur note comput pp damasio logic program system nonmonoton autom apt acycl gener aravindan ration algorithm view delet intern symposium logic mit pp abduct intern confer logic mit pp gottlob enhanc symbol model check ai research report institut fur universitat negat gallair minker logic data plenum pp relationship abduct journal logic role abduct view journal intellig inform damasio abduct extend logic nerod truszczynski intern confer logic program nonmonoton lectur note artifici pp inou sakama comput extend abduct extens sld abduct integr mainten view updat deduct joint intern confer symposium logic mit pp deneck de abduct procedur abduct logic journal logic grant model theoret approach view updat deduct journal autom fung procedur abduct logic journal logic gelfond stabl model semant logic intern confer symposium logic mit pp gelfond classic negat logic program disjunct gener lobo view updat stratifi disjunct journal autom updat knowledg base gener inou abduct framework nonmonoton theori intern joint confer artifici morgan pp inou fixpoint abduct logic journal logic inou transact extend intern confer principl knowledg morgan pp kowalski abduct logic journal logic kaka updat intern confer larg morgan pp abduct versu closur causal artifici abduct theori artifici brewka principl knowledg csli public pp lavrac induct logic techniqu elli marek specif mean european confer logic lectur note artifici pp logic framework default artifici theori diagnosi artifici rossi contribut view updat intern confer logic mit pp semant quantit logic ieee symposium logic ieee comput societi pp van ross semant gener logic journal 
DB/1388 3 applic learn italian languag depart comput engin school electr engin pob yugoslavia web engin tune advanc web educ internet develop technolog tool enabl interact creativ manag domain includ interfac artifici intellig base tool improv learn applic learn italian languag student product develop cooper paper present basic element applic implement discuss possibl gener environ describ experi explain lesson learn stress statist analysi success product reli classic approach introduct rapid growth internet medium internet technolog led point educ detach human book knowledg earli internet exploit educ institut dissemin research knowledg gener shape requir lot attent user sourc knowledg includ resourc inform drove user main subject step educ uniform page text imag context sensit system adapt hypermedia system attempt tune simplist fashion side common classroom teacher face row student student listen carefulli teacher learn juli august depart italian languag literatur school yugoslavia dipartimento di ingegneria ed ingegneria universita itali pose question bore subject postpon lectur suitabl session tradit educ approach strength weak good collabor coordin student limit knowledg hard upgrad fail adapt student educ lot learn make huge amount inform period paper present good effort unit advantag tradit cyber educ paradigm problem statement goal studi exploit outcom train environ paper special cover applic field teach italian languag environ easi maintain intellig educ environ design detail consid primari issu design data multifacet manipul intellig incorpor decid reli techniqu broadli standard packag solut offer requir function platform support abstract object easili deploi regular web browser addit softwar hardwar tangibl form underli manag easi access object easi function upgrad environ meant wider object improv train learn reduct train sophist industri intellectu capit declin learner train steeper learn outcom project includ basic customiz user domain methodolog qualifi compet test final realiz complet solut directli perform train technolog tend evalu abil adequ effect transmit inform great deal experienti evid real learn involv recept case languag issu intend examin posit learn place brain technolog design evalu basi effect student brain exist solut train concern learn wider exist numer flavor exist solut primarili aim prototyp specif train skill major put materi make author parti interest predomin type onlin educ static virtual textbook fix set network web organ textbook multimedia put studi similar standard book textbook advantag resourc net make deepen subject drawback attent drift interest easi find interest virtual browser window type educ regard potenti train technolog industri corpor train significantli drop product cost improv product qualiti challeng set develop educ applic offer amount interact adapt adapt import web access great varieti user work home teacher colleagu side recogn assist exhibit interact kei condit make adapt strictli categor educ describ gener techniqu critic adapt filter system user find relev drop knowledg knowledg pool system offer function intellig train system adapt inform system person inform knowledg specif user access inform perform blend inform artifici intellig ai orient plain multimedia simul condit hinder sophist approach includ adapt hypermedia system base appli set util user order adapt content link hypermedia page individu user intellig tutor system knowledg teach strategi provid adapt individu learn tutor good approach suit gener train requir collabor cooper perform system locat ah implement techniqu satisfactori lack robust educ tool cyber knowledg resourc rapidli accumul distribut call creation environ percept importantli adjust individu user requir enabl technolog circumst internet comput commun capac compar accustom logic deliveri standard solut logic retriev inform learn environ distribut network publicli electron equip network major environ transcend cyberspac resembl tradit classroom larg desk face straight row desk teacher longer present facilit group work make student question resourc answer fundament paradigm inevit teacher formula formula offer depend analysi presenc distant train import memor knowledg skill approach freedom reliev time posit anim student collabor answer question unanticip design system meant prevent teacher react obviou student question uncertainti requir knowledg base incorpor kind faq messag board knowledg exchang lead synergist result system monitor suppli materi suitabl teacher concentr main plain deliv guid student unforeseen encourag student benefit result approach highli qualifi person divid larger number achiev level compet tradit train environ customari student pose question uncertainti clarif offer mean commun student teacher level chat forum icq develop vrml interfac student classroom featur enabl synergist gener main qualiti classic approach type inhabit administr provid concern put piec knowledg tool upload capabl text raster similar file type requir provid contribut content call content provid knowledg provid resembl book publish administr concern give access peopl decid pedagog polici administr role develop system learn process mean learn materi fit ambit make structur knowledg student model student made student learn session monitor expert model learn perform curriculum decid knowledg materi hint offer mean student individu decid question converg person socrat fingertip special attent user interfac specif learn session occupi rare avoid activ window hint highlight item singl materi present detail import achiev comfort provid imagin blend ui design interfac reli robust technolog extens markup languag extens stylesheet languag cascad style sheet dynam support java program languag web deliveri solut advantag user reli web browser wide distribut central user provid advanc function java develop control behavior element render element latest xsl draft visual present css state approach enabl develop innov environ incorpor tangibl benefit virtual tradit learn minim shortcom approach classic languag learn tool tradit languag learn tool publish video accessori disc contemporari textbook base modern achiev appli modern psycholog learn linguist theori languag acquisit appli modern approach author place tradit mention commun aim gain level commun compet modern base theoret made european languag expert relev base initi clear theoret practic instruct linguist expert european council level threshold level base similar linguist independ european mention tool repres complet defin system enabl student follow process learn acquir foreign languag didact process base standard unit scheme textbook intend group learn individu learn scheme justifi introduc term respect prioriti easier unknown essenc approach solut relat relat linguist discuss essenc technic innov demand free user burden familiar technolog prerequisit minimum version web browser access web browser packag support multimedia standard found java applet javascript client side improv qualiti learn materi set static graphic raw audio file standard typic insuffici educ includ multimedia browser reli helper applic quicktim movi current effort main player web browser suppli product util automat download updat netscap microsoft activ paragraph condit assumpt enabl system access net architectur open compon level compon dedic transfer compon middl layer load balanc criteria enhanc reason capabl cbr modul integr part plug plai configur interfac defin metadata set standard complianc forward compat issu greatli reduc futur mainten effort reengin exploit implement specif elimin differ javascript world wide web consortium program modul pure java tm document project complet de facto standard design document softwar system uml diagram system design lot understand export java code enabl greater number develop programm project assum knowledg provid knowledg essenc linguist innov internet presum individu select mention possibl educ process train system tradit didact order relat tradit give possibl creat specif individu learn process approach languag learn presum initi possess level knowledg italian give result knowledg morpholog syntact categori knowledg basic terminolog detect wrong program lead kei lead grammat explan relat mistak option presum student updat knowledg creat automat individu didact curriculum student separ textbook vertic diagon tradit lose characterist tool type coher scheme tradit tool tool prospect greater shorter period lost time repetit teach program possibl creat group level acquir languag skill group creat program detect mistak particip subscrib connect encourag detect tini detail situat tradit languag teach difficult technic detail interest system present sketch architectur design view method outlin give detail architectur initi research encompass client applic server tier idea client thin perform busi logic java chosen languag offer test hold java remot method invoc client middl tier implement java http call pass argument applic server tier process input java connect commun jdbc driver implement api implement client inhabit middl tier resid data center campu compani inform server center ad architectur enhanc order fulfil extra tier server tier client applic reusabl servic server tier resid close reduc commun reduc cach prefetch mechan client http request process java webserv decid action servic server servic determin servic local servic rmi corba call remot server special featur knowledg provid curriculum sequenc author knowledg complex user implement java applic result architectur displai figur horizont number instanc comput commun client access server modem serv app server queri link optic link multimedia rich knowledg altern resid separ applic data replic mechan requir tier includ separ knowledg defin knowledg textual multimedia curriculum sequenc roadmap learn user profil complet user model includ prefer portfolio part expert model definit teach strategi commun applet creat rmi object regist object rmi object registri secur client invok java webserv url java servlet api manag secur addit tier solv problem reduc download time dynam download runtim applic cach applet static applic data closer client sandbox secur model java prohibit applet access local network resourc commun machin applet download sign applet relax secur set trust applet problem browser support trust applet substanti infrastructur ad acoust approach achiev import dimens compon ad system enabl student listen nativ speaker italian click individu student obtain acoust student practic skill listen client server applic tier receiv data request perform complex data acquisit issu set queri type architectur emploi wan environ corpor academ connect user java servlet receiv http rmi request determin servic user id pass augment javascript code connect user open creat cooki fast user profil activ user local perform inform exist object jvm memori object remain protect jvm memori updat long user drop queue environ optim approach consid java applet optim java tm perform engin applic pursu maximum java good employ applic consid applic intellig includ design architectur design program model program model tm compon design base outlin explan distinct issu provid section knowledg repres jdbc applic server figur architectur adapt individu advanc featur intellig cach adapt load balanc mechan implement applet resid server crucial compon intellig tutor system modul linguist experi mind train system successfulli appli industri thought program appli successfulli field languag learn acquisit chosen italian languag level possibl check immedi effect popul year student major italian school morpholog syntact problem italian includ element plane level knowledg curriculum aim enabl user gain adequ linguist compet special stress enabl read write speak italian translat bilingu dictionari properli consist part part consist didact label didact unit deal italian phonem orthograph didact unit deal central problem italian problem introduc italian text italian text gradual simpl text contemporari italian literatur acoust specif gradual lexic text literari written select order respond specif possibl student level knowledg text intend check read enabl lexic enrich applic acquir knowledg didact text popular contemporari italian song support acoust part text written section final examin consist translat treat case complement lexic grammat part text chosen read origin italian text prepar student autonom read linguist analysi contemporari italian literatur part contrast lexic point dealt part subpart subpart consist origin text acquisit check fix acquir knowledg instruct italian aim languag italian part newli introduc explain descript italian synonym knowledg level requir curriculum year student explan mean context cover statist student inform found internet teach process standard lesson teacher student possibl autonom internet occas written section questionnair distribut student final mark student decid oral fifteen dai written section exam valu correl questionnair made dai dai start internet major hypothesi underli questionnair summar internet grade final question improv internet attent student tire internet bring effect younger popul linguist questionnair order concret valu relat applic system field italian languag prepar questionnair question gener divid main wai possibl internet comput technolog wai possibl educ program italian languag learn question questionnair relat research hypothesi internet home home internet chat internet studi italian cultur student dictionari encyclopedia unreach greatest extent explan attract part grammat explan topic subject field italian civil close connect morpholog transform consid part close connect morpholog transform consid tire part student enrich greater extent authent materi topic subject field italian civil high school experi set textbook select elementari high school port high school student master subject creat statist compar experi expect posit effect technolog pronounc case younger popul detail experi scope paper continu educ experi popular textbook enabl gener popul learn italian port system book professor famou yugoslavia experi relat textbook creat end previou discuss essenti issu section discuss lesson learn find gener distribut student attempt internet technolog student high creat reliabl statist answer number import discuss student internet prepar percentag time spent compar total studi maximum figur tradit method deepli built habit typic student popul issu import distribut time spent student tendenc spend time long time expect total time distribut time chang futur expect futur student hope student spent time grade notic regular notic higher import exam question student achiev higher percentag answer case fairli situat classic approach question classic good thing expect student make easier follow materi internet technolog dramat percentag feel tire peopl fun student save amount time time rang save achiev advanc technic featur incorpor analysi student suggest domain subject follow research student habit interest feel technic structur function made compat gener habit interest innov effect system maxim analysi issu subject paper research left practic implic paradigm evalu practic revolut work attempt direct numer approach applic mean util technolog lot space research social implic suffici recogn good evalu experi show approach illustr futur effort research area cite object intern forum educ technolog societi encourag discuss issu affect educ system develop educ commun conduct multipl discuss thread specif topic approach develop specif aspect design implement integr learn environ sharpen vision process educ discuss aim definit reach discuss structur form problem fix term discuss end put concret form public blend good practic tradit exist train environ kei success train environ examin learn teach tendenc environ led creation hybrid train open futur develop insight internet domain make design crucial make system extens user futur develop forward creativ input insid environ futur plan includ prove concept util field studi research incorpor adapt ai techniqu expect follow environ win battl ultim user integr interfac linguist pedagog point demonstr student hold attent averag longer teach tool score exam tradit score exam effect pronounc student tire long hour learn note italian materi url chang interest parti consult updat refer divid refer paragraph found engin linguist engin refer brusilovski educ system review comput interact carnegi mellon java sun architectur applic internet vol number java bean ibm san francisco ibm white paper markup languag version web standard publish world wide web consortium object model level document version recommend stylesheet languag specif world wide web consortium work draft apr style level world wide web consortium recommend dec jan languag specif uml model languag object manag group intern forum educ technolog linguist refer italian il za 
Agents/3095 1 autonom fl commun reactiv multiag robot system tucker balch ronald arkin mobil robot colleg georgia institut georgia usa receiv abstract multipl cooper robot complet task quickli reliabl robot commun robot multipli capabl import commun robot societi investig experi simul real robot perform measur type commun task level commun progress complex potenti implement commun significantli improv commun appar unnecessari case commun lowest level commun effect complex type bulk result deriv thousand simul run randomli gener initi condit simul result determin paramet reactiv control system port test den mobil robot multiag mobil social commun introduct robot system design carefulli compon design addit robot justifi contribut effici task complet compon directli contribut add cost benefit commun compon multiag robot system merit care question simpli includ complex structur design formal methodolog design answer question georgia tech mobil robot robot system design methodolog develop refin singl multiag robot system system implement simul mobil robot approach reli kei object metric system iter cycl simul instanti real system design quickli discov control paramet critic paramet vari perform measur compar configur goal find system maxim perform metric configur port real robot system test approach appli commun reactiv multiag robot system discov commun impact multiag robot system societ robot task perform simul team robot measur task type commun experi design perform type commun compar task space control paramet explor commun number number mass percentag balch arkin obstacl coverag simul result support port control system team den mobil robot relat work multiag robot system multiag robot system constitut activ area research larg bodi literatur exist system rang size thousand robot dudek al provid taxonomi system classifi dimens group process commun topolog bandwidth research articl concentr small group typic order ten agent larg scale swarm robot system consid fukuda studi multiag robot system context refer cellular robot pioneer work concern heterogen agent research report articl homogen agent function research mit ai laboratori studi aspect reactiv control robot societi consist agent learn method evalu applic multiag system investig militari environ unit state europ planetari explor target domain societi research forag consist search environ object carri back central locat consum requir robot perform work attractor carri back similar robot robot team adequ cover environ forag wide studi date floreano describ forag strategi neural network architectur drogoul ferber report result simul forag robot demonstr spontan evolut structur chain extrem simpl agent press research articl role commun multiag robot system arkin report success behavior occur absenc commun agent creat societi consist group small robot conduct search retriev task recruit signal author report approxim improv perform target acquisit type signal work report preliminari werner dyer studi evolut commun synthet agent demonstr direct mate signal evolv system presenc societ necess studi problem conclud commun evolv societi simpl robot agent societi evolv commun fitter commun suppress order magnitud perform observ learn introduc franklin simul research conduct cooper multiag system studi role infer relationship lead special categori cooper system recogn inform explicitli request receiv order potenti multiag system studi commun specif context robot system research task defin requir commun coordin robot limit vocabulari selforgan time improv perform involv mimick behavior leader robot describ coordin protocol basi encod commun reactiv multiag robot system signal robot navig task formal theoret method appli limit problem wang distribut mutual techniqu coordin system research report articl motiv desir creat design methodolog multiag reactiv robot system effect design system import correctli number agent commun mechan robot societi task goal studi report biolog system natur offer wealth exist success behavior robot design directli appli work commun import natur societi inspir strategi creat multiag system significantli influenc biolog etholog studi report dimens commun system specif exampl role commun anim societi report commonli studi social biolog system ant excel refer social organ commun method includ ant typic chemic commun convei inform goss al studi forag behavior creat comput model capabl replic speci perform task frank behavior armi ant context group retriev prei relationship mass object retriev veloc return influenti work social behavior anim describ rang behavior simpl social cooper involv induct thing reciproc behavior feed antagonist mate behavior involv famili group life behavior involv commun attack herd behavior behavior involv reproduct fight mutual hostil group interest studi show impact forag behavior fish present factor consid includ food danger competit behavior commun understand emerg organ multipl agent natur commun support group behavior studi primat conduct organ coloni rel environ research displai behavior anim insight relat commun mechan articl task robot societi task robot system perform dictat extent sensor actuat requir appar task impact control system commun paramet investig gener multiag task task robot wander environ item interest encount robot move final attach robot return object home base ant speci perform task gather food robot perform task potenti suitabl garbag collect collect hazard environ figur show simul robot forag attractor return home base simul environ section balch arkin shown larg black attractor repres small path robot shown solid dash line leav dash line solid line return attractor home base mass attractor item dictat quickli robot carri heavier slower speed robot cooper move attractor maximum speed individu robot consum consum task involv wander environ find attractor encount robot move attach object unlik robot perform work object place attach time requir work proport mass object robot carri object back home base applic includ wast clean task figur show simul robot consum attractor note task perform environ forag task shown figur robot leav dash line solid line acquir move attractor mass attractor item dictat quickli robot consum heavier time robot cooper consum attractor faster task rate consumpt linear number robot ceil task differ consum discret attractor involv object complet visit environ familiar exampl sow cow task robot search area move entir environ percentag cover assum robot possess mean fix size task dictat proport environ cover complet figur show simul robot environ robot leav dash line solid line robot seed vacuum remov produc forest size robot percentag area robot affect long complet task multipl robot complet task faster avoid area find area quickli task paramet task definit includ paramet affect speed robot system carri ffl number attractor number attractor robot collect consum affect long accomplish task ffl mass attractor gener attractor mass thought factor factor consum task ffl coverag total size area percentag requir directli impact time cover section report result factor affect perform complex task basic task behavior robot perform consid result task import complex task easili combin simpler robot remov commun reactiv multiag robot system forag consum fig simul robot attractor work period return station remov portion task analog similar consum complex movement tactic util armi scout emploi group ground safe penetr hostil area group move forward short wait group move forward behavior perform built special coordin consum task waypoint group virtual attractor behavior emerg element group success move attractor attractor research laboratori underwai investig complex behavior combin basic behavior research includ languag individu societi robot formal formal oper behavior group complex assemblag assemblag combin form behavior robot languag includ oper coordin individu robot cooper group articl describ robot behavior formal relat philosophi appli reactiv control reactiv control system research provid reader summari reactiv control special characterist system reactiv control paradigm emerg approach control robot perceiv problem earlier research requir heavi relianc intern world model reactiv control distinct ffl multipl parallel behavior construct modular fashion ffl design system increment ad compet robot ffl percept action tightli coupl ffl relianc explicit world model knowledg avoid execut ffl dynam unstructur domain brook subsumpt architectur wellknown control paradigm repres exampl includ strategi differ wai includ organ decomposit behavior concurr process interest reader refer complet review reactiv control wide success laboratori simul studi real robot implement balch arkin featur distinguish robot control reactiv approach ffl dynam network process hardwir layer system ffl arbitr behavior execut concurr ffl potenti field techniqu encod robot behavior ffl flexibl introduc knowledg plan select system ffl adapt learn facilit flexibl ffl etholog studi provid motiv schema activ behavior comput reaction perceptu stimuli method analog potenti field note unlik tradit potenti field entir field robot reaction current percept world present locat independ comput sum normal robot execut cycl repeat rapidli facilit percept permit inform process basi problem local cyclic behavior potenti field strategi handl method inject system resort highlevel plan visit local continu adapt learn strategi robot control demonstr provid robust navig complex dynam world appendix inform specif comput individu schema research baselin assemblag paramet result gener task section compar perform robot system perform result baselin data comput select reason set control run statist number simul valu paramet base previou research behavior execut task baselin paramet highest task assemblag repres finit state acceptor consist state provid easi mean express reason behavior set provid formal semant state correspond separ assemblag constitu set motor schema instanti state activ transit state activ motor schema perceptu schema provid inform robot interact environ forag forag robot deliv robot begin wander state attractor robot field robot remain wander encount attractor transit acquir state trigger acquir robot move attractor suffici attach trigger robot attach attractor deliv state robot carri attractor back home base reach home robot deposit attractor revert back wander state figur show forag activ schema paramet ffl wander state high moder persist cover wide area environ suffici high avoid commun reactiv multiag robot system robot moder high forc individu robot effici cover environ perceptu schema trigger acquir state robot attractor ffl acquir state low deal local minima suffici high avoid low robot converg attractor avoid collid high gain move robot detect attractor perceptu schema trigger state transit deliv robot close attach attractor ffl deliv state low gain deal local minima suffici high avoid acquir high home base target perceptu schema trigger state chang robot reach home base specif valu schema gain paramet studi list tabl deliv acquir encount deposit attach wander fig forag consum behavior consum task similar forag schema gain ident wander acquir state consum uniqu behavior consum motor activ reduc mass attractor fix rate time attractor fulli consum deactiv robot transit back wander state tabl show schema paramet consum wander acquir state similar forag consum primari differ wander state replac similar schema fix sensor rang detect area attractor robot start wander state search area encount transit acquir state move robot arriv transit state state state robot move current head fix environ long continu area directli robot remain state activ schema state ffl low deal local minima acquir encount attach consum complet wander fig consum balch arkin acquir encount complet arriv wander fig ffl high avoid ffl robot close avoid ffl probe moder encourag robot move current head area ffl perform actual oper fix ffl perceptu schema trigger state chang robot complet local area simul implement maintain mark high resolut grid environ entir grid mark robot mark visit area grid gain paramet schema activ state list tabl form commun type commun evalu research minimalist type involv direct commun agent type transmiss state inform agent manner similar found displai behavior anim type requir transmit agent recogn broadcast locat attractor locat detect rang form commun detail commun type multiag societi direct commun robot discrimin intern perceptu obstacl commun agent robot reli percept world arkin shown previou work basic inform support cooper robot retriev task cooper context refer observ phenomena multipl agent converg work task baselin result show cooper emerg consum task state commun state commun robot detect intern state robot result report commun simpler bit data indic agent wander state indic state wander type commun shown provid distinct advantag commun perform forag task commun consid state commun necessarili inform relai passiv observ sender necessarili explicitli broadcast observ natur type commun demonstr anim postur extern dog exhibit flight behavior fear advantag state inform reactiv behavior assemblag task modifi slightli robot point import state robot robot found attractor area mean robot found work robot find commun reactiv multiag robot system work assist cooper state wander state robot found work behavior modifi robot transit acquir discov robot robot locat robot attractor detect head directli goal commun goal commun involv transmiss recept specif inform implement mobil robot requir data decod goal commun differ level sender send broadcast inform natur type commun found behavior bee discov rich sourc return hive commun locat encod direct distanc hive sourc reactiv goal commun implement modifi behavior assemblag manner state commun transmit robot discov receiv robot move directli locat attractor intent agent follow direct path attractor rudimentari form commun broadcast goal transmit agent involv mode involv transmiss detect attractor independ transmit agent acquir deliv present option receiv move closest attractor independ transmit agent benefit addit form commun left futur work explicit versu implicit commun implement goal state commun requir explicit signal recept commun inform state commun implement simpli mount binari signal atop robot depend robot intern state explicit requir act invok signal inform pertin cooper gather mean intern state robot infer observ movement recogn robot wander state due appar random place larger perceptu burden receiv agent robot commun environ robot leav evid passag place visit modifi fact observ robot type commun refer implicit requir act transmiss implicit commun found import mode cooper simul task commun emerg interact agent compar perform implicit commun simul environ simul environ provid accur estim robot perform real world simul import offer mean test robot system configur quickli simul report perform term prescrib perform metric realist emul environ robot interact simul variabl readili manipul test environ research written window graphic packag simul tool research mobil robot lab georgia balch arkin includ result gener simul environ routin demonstr actual mobil robot minor present simul earlier project simul run visual mode visual mode primarili debug qualit access mode multipl run gather extens statist data robot ident vehicl control task assemblag agent current depend sole percept robot agent execut task unit environ unit conveni comparison real robot implement repres foot time measur step step iter program calcul robot posit robot locat detect attractor robot fix radiu field view grasp carri consum task dictat simul automat enforc limit rule set task limit robot move restrict move perform metric goal research report impact commun robot perform object measur select perform metric import metric competit cost versu reliabl potenti metric multiag robot system ffl cost build system accomplish task minimum cost industri task metric tend reduc cost system minim number robot ffl time build system accomplish task minimum time metric lead solut call maximum number robot oper ffl energi complet task smallest amount energi situat energi store space applic ffl build system greatest probabl complet task time cost strateg militari applic task metric numer combin measur metric simul time complet task chosen primari perform metric easili accur measur conform frequent thought perform claim made robot path length energi consumpt equal simul studi perform measur count iter simul program execut task complet initi condit task prevent robot complet attractor circl robot reach scenario solvabl robot system capac move obstacl ultim potenti defeat pure reactiv strategi present provid simul continu step failur declar run complet highli system complet task failur declar object evalu impact commun make import system measur improv commun case run record step approach report optimist perform run complet show improv failur system commun reactiv multiag robot system complet task step factor target system oper environ incorpor design process control system factor includ simul import factor ffl mobil terrain mountain percent environ serv ffl obstacl percent environ clutter ffl metric priori robot good map area complet ffl static environ fill move reduc util environ static static flat environ randomli scatter obstacl assum priori knowledg obstacl locat obstacl coverag vari total baselin motor sensor constraint step robot system design realist bound expect motor sensor capabl robot set bound reduc search space optimum solut affect commun perform main thrust fix valu repres expect perform robot goal determin optim sensor motor paramet vari tabl show motor sensor valu simul baselin result analysi tool build baselin perform configur task paramet select empir baselin serv control comparison evalu commun experi gener run simul baselin configur paramet number robot number attractor object percentag vari combin robot measur perform time run randomli gener scenario perform averag run simul record number step run baselin perform measur made commun robot control compar perform task state goal commun quantit mode commun impact perform basic perform perform data visual surfac axi reflect number robot axi indic number attractor percent coverag figur axi show averag time complet task combin robot attractor number indic plot task share similar shape notic back left corner highest point surfac expect locat repres case robot complet work attractor forag coverag front lowest largest number robot complet amount work appar task perform initi improv sharpli robot taper perform improv balch arkin tabl paramet valu note valu task factor baselin rang task factor number attractor mass attractor avg coverag factor obstacl coverag obstacl radiu number robot sensor motor constraint maximum veloc fix attractor sensor rang ft fix obstacl sensor rang ft fix commun rang ft fix commun type ft fix consum rate fix control paramet obstacl sphere influenc ft fix obstacl gain fix robot sphere ft fix robot gain fix robot gain fix robot gain fix gain fix gain fix probe gain fix robot import robot robot system forag task fast fig optim time cost perform defin time complet task number robot time perform combin time complet task cost system design balanc import cost versu speed approach amort cost robot system expect lifetim cost run cost divid expect number run amort cost robot run valu time step number time complet perform time measur ad amort surfac gener perform metric system robot gener attractor environ expect commun reactiv multiag robot system fig time complet task robot attractor commun robot choic robot goal design call fewer speedup effect tool speedup measur plot speedup reveal effici robot complet task perform robot speedup point robot complet task fast speedup number matar introduc similar metric robot perform speedup equal perform linear superlinear perform greater sublinear case robot faster actual task complet offer sublinear speedup figur show speedup plot commun note speedup task gener higher larger number attractor research branch comput scienc found random search task complet superlinear time parallel system wander behavior task essenti solv random search perform superlinear behavior heavili case larg number attractor speedup consum task sublinear point behavior consum state offer linear speedup limit set specif environ massiv attractor forc speedup limit fig speedup task robot attractor balch arkin tabl summari speedup data task task avg speedup worst forag consum hypothesi test reduc averag mass rerun simul baselin attractor mass vari mass reduc unit reduc attractor mass robot spend time wander superlinear consum speedup consum lower mass attractor shown figur point speedup low mass attractor high mass case speedup superlinear speedup task superlinear point surfac worst speedup dip situat requir high percentag coverag result peak robot coverag case high coverag robot spend time wander bit area wander superlinear time speedup expect regim speedup result summar tabl fig side side comparison speedup consum task perform attractor averag mass timeout timeout occur simul run exce time limit limit timeout mechan avoid infinit loop event societi unabl complet task world frequenc timeout combin robot attractor measur plot figur frequenc timeout serv primarili measur data qualiti situat timeout frequenc long run complet run complet run indefinit perform greater certainti timeout occur fewer robot solv task attractor higher coverag requir summari baselin result baselin result serv control comparison assess impact commun mode perform import deriv understand fulli basic result test complex robot configur import result baselin configur ffl number robot complet task faster fewer robot ffl number longer complet task attractor commun reactiv multiag robot system fig frequenc timeout task robot attractor ffl perform metric result system optim lower number robot metric ffl speedup greater scenario larger number attractor present ffl speedup consum task superlinear lower mass attractor ffl speedup task superlinear ffl timeout occur low number robot high number attractor result commun commun task figur show typic simul run robot forag attractor goal commun inspect imag left reveal appar improv robot path quantit result summar tabl confirm qualit impress figur show typic perform plot case commun perform data point repres simul run plot goal commun similar contour improv perform evidenc lower surfac commun complex statist analysi tabl summar observ quantifi differ perform perform ratio plot comput perform commun divid perform commun result greater impli improv perform improv case state commun improv perform task averag goal commun state commun task commun consum task impact commun perform consum task similar figur show typic simul robot consum attractor goal commun result simul goal commun longer state commun slight run time goal versu state commun typic task repres basic perform data simul consum task plot figur contour form commun similar compar analysi reveal state commun offer perform advantag commun goal commun averag state commun goal commun tabl summar result balch arkin commun state commun goal commun fig typic run task state goal commun simul requir complet fig perform ratio plot task goal versu state commun recal speedup consum task link attractor mass attractor mass impact benefit commun analysi data run low mass attractor reveal goal commun perform indistinguish state commun futur research determin result anomali task paramet shift trend commun task result task simul commun plot basic perform data level commun shown visual ident figur case state commun commun perform goal commun virtual indistinguish state commun tabl summar result robot inevit leav record physic chang environ form implicit commun robot leav mark work complet result import impli task implicit commun explicit commun unnecessari commun state commun goal commun fig consum task goal commun simul requir complet commun reactiv multiag robot system task averag improv worst forag state commun goal commun goal state commun consum state commun goal commun goal state goal state mass state commun goal commun goal state commun tabl summari perform ratio state goal commun summari result commun perform improv type commun offer task summar tabl import ffl commun improv perform significantli task implicit commun ffl commun unnecessari task implicit commun exist ffl complex commun strategi offer benefit basic commun task displai behavior rich commun result mobil robot ultim goal research work multiag robot simul serv develop tool demonstr simul move complet function behavior instanti mobil robot target system group den mobil kinemat ring ultrason rang sensor ren initi result obtain port task motor reactiv control system written omnidirect behavior benefit omnidirect sensor enabl robot discrimin attractor obstacl omnidirect sensor system evalu multiag robot implement georgia conic mirror camera laser barcod read system conic mirror camera system origin develop mackenzi georgia tech unman aerial vehicl team intend local unman robot helicopt bright light locat environ serv refer point triangul chief advantag imag process demand compar low conic vision readili adapt mobil robot system marker light attractor robot robot light arrang attractor typic configur shown figur system atop robot ren figur figur actual imag system attractor robot visibl short bright blob left side imag azimuth attractor comput robot equip light azimuth rang balch arkin attractor camera robot conic mirror fig typic scene conic vision process fig imag conic vision system bright blob identifi attractor robot environ comput figur show separ blob result light robot grow approach laser barcod reader manufactur den mobil robot laser posit center devic rotat mirror inclin scan sensor detect accur report azimuth target code tape feet barcod reader shown ren figur barcod reader support driver reactiv control util local attractor detect task system suit multiag robot research present barcod reader georgia imposs conic vision system support driver omnidirect capabl ren shortli laboratori acquir sensor ir laser rang scanner task section port test ren requir schema code lack exist omnidirect sensor system attractor robot detect complic matter problem circumv simul sensor perceptu schema util shaft encod data spatial locat attractor move robot maintain continu updat share file fidel maintain code perceptu schema locat attractor robot sensor rang test depict figur sequenc shown videotap imag captur print tape telemetri run shown figur ren set home base attractor mark circl floor foreground background inact left involv task schema robot robot activ ren sensor paramet set baselin simul test area attractor sensor rang reduc feet prevent ren immedi attractor home base fig imag robot drawn closer note separ blob commun reactiv multiag robot system attractor attractor fig den demonstr task ren tag attractor telemetri forag demonstr shown home base center attractor locat left inact robot lower left foreground attractor feet home start test immedi visibl ren move figur notion attractor carri back home base ren transit wander note locat ren remain attractor wander includ strong robot robot ren continu search attractor assumpt search rest inact commun attractor inordin amount time discov human step human herd ren attractor place hand ultrason sensor ren feet transit acquir human robot tag attractor deposit home base robot run task shown figur paramet baselin simul attractor sensor rang set feet minimum rang robot approach obstacl set feet attractor obstacl environ robot initi home base run made commun begin run robot enter wander immedi detect separ attractor tag respect robot deliv home base robot cycl wander attractor remain attractor ren sensor ren approach ren return attractor home carri sensor rang respond approach ren deliv attractor reconstruct run shown figur commun mode consum level commun consum task implement test ren scenario robot attractor test consum behavior scenario simpl serv illustr advantag qualit differ level commun section run mobil robot directli compar simul scenario figur test robot attractor arrang robot immedi sensor rang sensor rang attractor feet lower robot commun robot initi move attractor robot move due commun robot initi move attractor balch arkin fig den ren demonstr forag task ren tag attractor attractor ren deliv attractor home base predict born simul shown top row figur simul run environ section baselin control paramet case robot immedi move attractor begin consum robot move continu search attractor wander state eventu fall sensor rang move consum case state commun robot initi move attractor robot begin follow transit acquir state sensor rang attractor case goal commun robot immedi move attractor consum qualit differ state goal commun visibl path robot attractor figur state robot initi sensor rang make curv path attractor follow robot initi goal commun robot proce directli attractor compar simul run robot ren sensor rang robot set scenario alter run mobil robot attractor feet lower robot telemetri shown half scale simul run account smaller scale scenario perform mobil robot commun similar simul perform bottom ren attractor explor left side laboratori sensor rang move attractor state commun ren make curv path commun reactiv multiag robot system home base attractor obstacl attractor attractor ren home base attractor obstacl attractor attractor home base attractor obstacl fig reconstruct forag demonstr goal commun ren travel directli attractor path lower robot case state goal commun simul mobil robot mobil lower robot curv upper robot simul result factor scale telemetri half simul effect visual perceptu process obstacl detect ring ultrason sophist ignor robot detect robot obstacl problem resolv omnidirect sensor perceptu process incorpor research summari impact commun perform reactiv multiag robot system investig extens simul studi perform result gener task illustr task environ affect commun payoff initi result test mobil robot shown support simul studi princip result task ffl commun improv perform significantli task commun ffl commun essenti task includ implicit commun ffl complex commun strategi offer benefit commun detail section articl futur work involv major research thrust concern societ perform multiag robot commun present robot agent potenti failur research thrust involv integr human effect control societi area includ develop method formal express multiag robot system goal produc tool facilit establish formal provabl properti suffici specif appendix motor schema formula appendix method individu primit schema research comput compon vector result activ schema sum normal prior transmiss robot execut ffl attract goal variabl gain set high head goal adjust gain direct direct perceiv goal balch arkin commun state commun goal commun robot robot attractor robot robot attractor robot robot attractor ren attractor ren attractor ren attractor fig comparison simul consum task run run mobil robot ffl repel object variabl gain sphere influenc avoid adjust sphere influenc extent forc center radiu obstacl adjust gain distanc robot center obstacl direct line robot center obstacl move obstacl ffl random wander variabl gain persist overcom local explor adjust gain direct random direct persist persist step persist ffl favor continu motion current direct head adjust gain area detect direct straight ahead extrapol path current locat area ahead direct import area ahead gain acknowledg research support nation scienc foundat grant grant mobil robot laboratori support addit grant nsf river technolog center author doug mackenzi russ clark develop softwar project commun reactiv multiag robot system note task earlier state mention correspond state move object schema escap move object differ test scenario robot fail complet restrict robot movement obstacl overlap result accur account obstacl coverag research robot initi center home base earlier research randomli environ author singl start locat robot real world implement previou simul run exce maximum step run time limit halt discard limit step run timeout count time step robot move visual boundari case percent area vari increment difficulti vari discret step result directli compar forag consum task attractor refer initi result mobil robot work note workshop dynam interact robot pp energi american motor schema base mobil robot intern journal robot vol pp impact cybernet design mobil robot case studi ieee transact vol pp cooper schema base robot journal robot vol pp integr world knowledg reactiv navig design autonom ed pp model neural function schema implic result robot control biolog neural network ed academ pp surviv robot reactiv control robot remot system hazard ed pp commun behavior state retriev proc ieee intern confer robot vol rc dimens commun social organ multiag robot anim animat proc intern confer simul adapt dec mit pp rc coordin perceptu algorithm mobil robot ieee transact robot autom mobil robot dock oper manufactur progress visual perceptu proc ieee intern workshop intellig robot system pp al instanti reactiv robot proc intern confer intellig autonom pp avoid simpl effect strategi reactiv proc ieee intern confer robot vol pp robust layer control system mobil ieee journal robot base construct ieee intern workshop intellig robot system japan pp learn perform enhanc reactiv system proc ieee intern confer robot pp effect food competit choic forag locat anim vol pp taxonomi swarm robot proc intern confer intellig robot system pp tom thumb experi forag balch arkin robot anim proc intern confer simul adapt mit pp emerg forag strategi ecosystem neural network anim proc intern confer simul adapt mit pp rf element cooper behavior intern research develop final report research institut michigan ann mi team social group retriev pre armi ant behav ecol structur robot base cell structur ieee intern confer robot pp ant michigan press trail lai trail solv forag problem ant coloni behavior mechan food ed rn nato asi vol pp sensor swarm ieee intern confer robot pp model greater ieee transact parallel distribut vol pp cambridg mass plan design autonom mit pp social etholog south wale anim behavior obstacl avoid manipul mobil proc ieee int conf robot st gener potenti field approach obstacl avoid sme ri technic paper implement procedur reason system multirobot applic confer intellig robot space autonom helicopt posit determin onboard integr vision proc sme appli vision confer ga formal specif proc spie confer mobil robot pp dc balch make clean behavior base vacuum work note aaai fall instanti agent nc synthet approach studi commun artifici life sfi studi scienc vol ed farmer situat agent design autonom mit pp minim complex control mobil robot popul proc ieee intern confer robot multipl planetari surfac mission proc ieee intern confer lo novemb pp disappear replac displai journal theoret strategi coordin mobil proc intern confer intellig robot system pp coordin approach formal coordin mobil robot proc intern confer intellig robot system pp adapt action select cooper agent team anim proc intern confer simul adapt mit pp intern action design autonom mit pp learn reactiv control paramet genet proc intern confer intellig robot system pp reactiv method select adapt reactiv control paramet autonom robot technic report colleg georgia tech commun reactiv multiag robot system driven local navig mobil jpl public jet ca social behavior london mutual base dynam proc ieee intern symposium intellig pp evolut commun artifici organ technic report ai lo angel adapt commun protocol cooper mobil robot anim proc intern confer simul adapt mit pp tucker balch born florida receiv bs georgia tech ms uc davi pursu phd autonom robot georgia tech support research lawrenc livermor nation laboratori comput scientist enter air forc pilot candid complet fighter train fli georgia air nation guard research interest includ integr delib plan reactiv commun parallel algorithm robot navig student member ieee acm ronald arkin associ professor director mobil robot laboratori colleg comput georgia institut technolog dr arkin research interest includ reactiv control percept navig mobil robot robot learn autonom cognit model percept technic public area fund sourc includ nation scienc river technolog offic naval research dr arkin associ editor ieee expert journal consciou member editori board autonom robot journal appli intellig senior member member aaai acm 
IR/1005 5 resourc discoveri knowledg discoveri internet zaian school comput scienc simon fraser canada abstract year time modern comput didnt exist bush wrote multimedia digit librari human collect knowledg fill link materi topic end world war urg scientist build knowledg store make continu extend access consult closest materi dream web hypertext multimedia document collect access knowledg realiz extens research accomplish inform retriev search final adopt text repositori system late advent web text search routin million user search engin daili pinpoint resourc internet resourc discoveri internet frustrat useless simpl keyword search convei hundr thousand document result knowledg discoveri data mine web research topic attract tremend interest realiz inform data knowledg resourc document web tabl content introduct inform retriev technolog brute forc document scan convent document retriev document cluster document signatur invert index term weight assign hypertext multimedia survei resourc discoveri internet internet tool inform retriev commun servic inform storag inform exchang servic inform index inform retriev servic interact multimedia inform deliveri servic catalogu directori robot search engin search engin work relev rank document retriev agent inform retriev multimedia resourc discoveri data mine knowledg discoveri internet web content mine web page summar search engin result summar web structur mine web usag mine refer appendix glossari inform retriev web introduct johann introduc print press europ centuri spawn revolut publish reproduct public enabl larg scale todai web bring revolut electron publish wit mass product public sort text sound program made singl moment overwhelm product resourc difficult access end world bush wrote articl invit scientist join effort massiv task build system hold sum human make store inform bush difficulti publish unduli view extent varieti public extend present abil make real hypothet machin scientist store thought link materi thought web technolog conveni author tool partial mimic bush descript todai million user continu contribut remark wealth knowledg issu bush remain solv bush continu todai web consent storag extrem larg amount multimedia continu updat make cumul find inform retriev effici effect found inform retriev fertil research field techniqu implement success prevail advent servic accumul aggreg inform retriev techniqu adapt bring index model search engin find document pertin request resourc discoveri resourc discoveri process clarifi inform retriev request identifi retriev resourc relev request function search engin perform effect tool satisfactori irrit annoi result current search engin technolog invit research tackl challeng index special inform gather filter cluster trend research inform gather internet knowledg discoveri knowledg discoveri internet process usenet ftp http text postscript bibtex html imag resourc discoveri knowledg discoveri format servic method capabl autonom learn agent figur internet space extract discov resourc global network knowledg discoveri internet larg inherit data mine domain access worldwid data mine relat internet goal resourc discoveri find explicit goal knowledg discoveri bring light implicit knowledg necessarili state resourc evolv space myriad servic emerg varieti resourc exchang protocol document internet progress simpl text fulli structur document multimedia record tool inform retriev skill start manual sift tool resourc tool learn exploit servic inform deduc knowledg document content tool classifi class tool direct resourc access class resourc discoveri tool search engin class class encompass guid autonom agent target goal retriev filter inform sophist agent fourth class agent abil adapt learn experi infer user user profil resourc search access pattern web mine tool discov extract knowledg found resourc constitut class etzioni compar evolut food chain organ agent feed search engin document peterson metaphor biologi taxonomi compar inform retriev tool ax figur repres document prevail axi retriev capabl method evolv inform retriev method pertain set perform type enjoi retriev power retriev method categor axi depend retriev resourc knowledg internet paper survei techniqu resourc discoveri internet trend knowledg discoveri field appli internet survei text retriev technolog document technolog variat adapt technolog heart resourc discoveri method internet inform retriev technolog librarian adopt comput technolog earli librari scienc research dealt problem manual index text retriev introduct comput technolog librari scienc permit autom index retriev librari catalogu autom bibliographi creat past work inform retriev librari autom document retriev import area patent law offic text busi offic text retriev document inform retriev technolog drawn attent research materi network commun brought challeng stimul inform retriev term coin time paper field subsum topic rang data retriev knowledg relat inform process task regard synonym document retriev focu text retriev aspect inform retriev field text document distinct search rank index refer repres document retriev search refer process examin document attribut document search queri final rank refer process order document retriev respect relev search queri section examin activ document retriev techniqu inform retriev system evalu base effect measur effect typic recal document retriev search document collect divid group consist document retriev document omit group subdivid document relev retriev retriev document retriev document left total collect relev relev document document collect collect recal figur recal ir relev search queri relev figur show tabl present group document defin recal good inform retriev system minim maxim fraction retriev document relev queri system maxim obviou evalu inform retriev system easi maxim simpli retriev document document relev case recal fraction actual set relev document correctli retriev system number relev document retriev relev document inform retriev system maxim recal obviou simpli retriev document collect maxim recal relev document necessarili collect recal maxim found recal measur select effect recal insensit total size measur reject effect fraction document retriev inform retriev system minim calcul recal requir knowledg total number relev document collect larg practic imposs number retriev relev document estim sampl techniqu method pool estim number import note recal depend definit relev relev pair rate individu relev rate obtain expert user measur highli subject repres user avoid popul user collect recal compar inform retriev system perform time perform system evalu brute forc document scan obviou document match queri string document search string brute forc method locat document exact redund redund search scan document method index document document space overhead lack scalabl full text scan method larg document algorithm string match base automata theori automaton singl regular express approach search search simultan straightforward algorithm substr test document larg string string search string compar charact point pointer increment pointer match reset string pointer set charact increment document string pointer search string slide document simplic slow size string size algorithm comparison morri pratt boyer moor algorithm perform comparison algorithm slide string increment algorithm compar string left variat algorithm recent sundai suggest variat mention full text scan method larg collect document lack special hardwar implement function program languag perl script languag import mention advantag method ineffici full text scan method requir space overhead invert time size document index updat document updat document necessit updat index rank document base weight calcul frequenc string document frequent search string relev document convent document retriev clear inform retriev research commun index kei success inform retriev system research devot lot attent develop autom index techniqu myriad techniqu implement test heart modern techniqu resourc discoveri internet outlin index differ significantli inform retriev method method index techniqu index structur index invert file sort term bit string constitut index inform retriev bit presenc absenc word document index word occur high frequenc elimin index process list call stopword consult word encount document queri commonli occur word discard queri submit natur languag mean reduc set word includ index plural convert singular languag normal techniqu borrow natur languag process reduc word similar mean gener common concept call conflat techniqu reduc word root singl form strip root deriv inflect word transform canon form english strip suffix suffic conflat techniqu languag depend base dictionari common word dictionari class word morpholog term stem extract word reduc term cluster synonymi class subsumpt group class word normal reduc size proven improv select effect inform retriev system inform retriev system natur languag techniqu word normal system notori elimin stopword articl pronoun index full text document content size index importantli recal system includ phrase phrase recognit phase process stopword normal phrase origin form altern phrase base syntact word manual built phrase dictionari combin approach inform retriev system involv document process record featur assist match queri index associ term document phrase gener document queri process similar manner extract featur document featur queri featur compar separ document relev queri order select document prepar relev system relev step involv user mark relev document displai resubmit rank system judgment system modifi origin queri difficult formul queri retriev system document collect relev feedback yield progress refin reformul queri refer feedback significantli improv retriev perform tradit inform retriev approach librari scienc signatur file method cluster index document gener bit string system popular todai hybrid variat technolog invert file index commonli approach inform retriev system method outlin subsect artifici method neural network genet algorithm cover survei document cluster idea document cluster similar document tend relev request group similar document search space reduc search document cluster involv cluster index cluster search queri process match cluster techniqu oper vector document repres number select index term document term chosen repres vector posit document vector absenc term document presenc term term weight occurr frequenc term document calcul term weight base rel occurr term vector model yield document space document point space partit point group constitut cluster format cluster format method classifi type iter hybrid method method base similar method appli graph theoret techniqu requir similar function measur close document relat document similar measur literatur method class basic connect edg point repres document document similar measur exce threshold document maxim cliqu connect result graph cluster number document method requir method graph theori requir quadrat time iter method requir similar function necessit empir determin threshold number cluster size iter method oper order log log gener approach determin initi partit iter document cluster assign improv cluster base measur method overlap cluster document cluster appli term term relat relev group term cluster reduct cluster repres concept vector term repres concept term reduc keyword dimens hybrid method combin similar matrix iter method salton present method iter approach gener rough partit document method subdivid partit smaller cluster cluster search start process queri repres vector cluster queri similar function requir function compar queri vector cluster centroid search cluster similar queri vector exce threshold vector model cluster method relev feedback relev document retriev mark system reformul queri vector restart cluster search effect search shown iter cluster techniqu attract interest lee present similar base approach solv data problem cluster method techniqu appli word classif natur languag process document cluster speech recognit reduct document signatur signatur file document retriev document yield bit string call signatur filter signatur document collect signatur file form index signatur store sequenti order document document store pointer document document signatur gener transform word document bit string concaten code perform similar transform queri word compar result bit string document determin document queri word smaller document signatur file search faster hash function method gener document retriev relev signatur file method effici distribut signatur uniform import select good effect hash function techniqu aim uniform distribut variat signatur file method base choic hash function approach adopt combin word signatur document signatur prove signatur probabl posit occur minim number document signatur equal number space wast number pair match bit convei optim half bit set bloom automat select size document signatur convei probabl bit filter size filter calcul base distribut number word document collect number transform perform word signatur file suggest tree structur document signatur cluster signatur approach improv search speed signatur file low accuraci rate signatur file method popular low space overload requir method major drawback entir signatur file consult method term disk oper number document larg reduc cost robert suggest store signatur file organ signatur file row signatur bit signatur file organ signatur column bit document signatur store consecut bit signatur row structur random access file reduc oper drastic updat easi normal signatur bit slice signatur file costli mainten time document updat signatur file chang invert index relat model model commerci invert index commerci inform retriev invert index invert file term set document list document file order alphabet order term term compound word phrase term store sort invert file pointer post file document access number document figur show term document pointer point record post file pda pointer store pointer store consecut post file point document document file structur easili built document collect cumul counter select term search process simpli locat document specif term retriev document pointer post file point term entri invert file tree structur top sort invert search process extrem fast invert file structur permit easi process boolean express set oper document set term express simpl intersect document set point term set point term yield document satisfi union set term invert file pda doc ptr post file size ptr list doc list doc document file pda figur invert index file structur yield document satisfi oper process manner problem invert file structur storag overhead invert file structur requir origin document collect size depend term chosen overlap synonym disadvantag method mainten cost ad document involv updat document list post ad term necessit costli tree structur updat easi implement speed invert file outweigh make method larg todai web search engin applic simpl structur invert file convei addit inform rank document retriev relev rank document base heurist number term queri popular term collect inform obtain invert file structur enhanc inform file structur addit file hold posit term document chang post file schema ad pointer post file figur show structur term document pointer invert file point record post file consecut record structur figur record post file pointer document number occurr term term invert file post file size ptr list doc list doc document file pda posit file pda ptr doc ptr po occur figur invert index file structur posit file document pointer list posit term document list store posit file time document pda posit number occurr term document relev rel posit search term document rank relev document proxim term close term retriev queri term weight assign term document equal content term import give weight term import document rank document relev term weight approach order assign high weight term deem import low weight import frequenc term document respect frequenc collect term frequent document document term repeat document import term document weight equal term frequenc time frequenc term collect definit weight favour term high frequenc document low frequenc collect document collect repres weight term vector form weight term document number distinct similar measur calcul pair vector measur reflect text similar document queri process manner weight vector queri similar comput form comput document collect similar rank list document order similar obtain queri sampl document hypertext multimedia major resourc internet resid web web hypertext web start hypertext import provid hypertext develop hypertext basic concept introduc ted coin term bush idea ted nelson develop project aim place entir world literari corpu hypertext text provid access inform incorpor notion tailor present hypertext defin approach inform manag data store network node connect link node sourc form hypertext call hypermedia major origin concept hypertext concept link link node document hypertext system collect node interconnect unidirect link permit backward document link capabl organ text figur show hypertext document node link origin call node link point call refer start point link refer anchor button read user present node consid document part convei theme click link user node user interfac paramount hypertext system navig larg amount inform author document system cognit refer cognit task schedul problem author addit effort concentr maintain task track datamin discov pattern trend inform retriev resourc internet manag system knowledg discoveri internet extract implicit inform network collect network internet figur hypertext document link user interfac issu hypermedia author issu drawn attent research commun major inform retriev approach model approach simpl navig document node node link major user tendenc lose locat direct document problem aspect problem address user interfac offer browser effort put issu hci ui research solut cope queri mechan node tradit inform standard inform retriev index techniqu appli usual queri boolean oper combin user locat interest node give addit inform rank heurist exploit node relev rank survei resourc discoveri internet moon land event time advent internet greater impact societi rememb centuri atom bomb centuri centuri birth global commun network atom bomb reaction politician emot impact societi mean rememb bad experi civil internet collect servic continu grow level imagin todai internet necess afford lose entertain internet start initi advanc research project cold war build depart explor possibl commun network surviv nuclear attack initi network connect research centr connect introduct commun protocol earli interconnect research result internet backbon nation scienc foundat establish supercomput centr unit state dedic network fund connect centr internet backbon growth number host growth packet traffic backbon tabl figur graph figur show growth internet host domain number web site number growth internet advent web connect commerci provid growth import number softwar tool make multitud resourc network transfer protocol adopt creat resourc internet file transfer protocol frequent servic internet term data packet subspac million document estim terabyt scatter thousand subspac usenet space web space contribut size internet interest note attempt made index aim internet subspac archi index ftp veronica index gopher search engin cover web recent web index includ usenet subspac ftp gopher internet tool inform retriev section outlin widespread network inform retriev tool servic number tool todai larg grow quickli classif order popular servic internet resourc retriev resourc categori network inform retriev tool servic categori encompass commun servic electron synchron chat categori group inform storag inform exchang inform index inform retriev categori interact multimedia inform deliveri web fourth categori commun servic electron mail main internet applic send messag user comput recipi user destin user machin connect network messag text busi memo person letter multimedia imag video major document gener servic privat contribut larg public resourc internet corpor archiv offic memo exchang archiv person electron mailbox grow rapidli necessit inform retriev date host domain web site figur growth internet figur growth internet term number uniqu host figur growth internet term number uniqu internet domain figur growth internet term number web site techniqu locat data mine techniqu summar messag content usenet usenet collect host receiv network discuss group forum varieti topic network mechan broadcast messag host larg number host messag articl receiv host usenet peopl access usenet host read messag connect simul messag broadcast peopl access group read post repli messag group messag archiv access thousand group topic scientif polit larg collect articl post daili usenet contribut significantli rapid size resourc accumul internet relai chat textual chat popular earli dai internet unix facil servic interact commun remot user irc session archiv busi chat session save corpor intranet retriev refer grow popular commun servic ad enhanc regularli newcom net net video significantli contribut larg number resourc figur timelin internet inform storag inform exchang servic anonym file transfer protocol access remot comput ftp server portion file directori structur exchang file web wide anonym ftp archiv site publicli access ftp wide access sourc inform internet ftp site offer file relat topic interest file document execut data file comput store ftp archiv site maintain volunt effort file directori access anonym explain content file directori standard content structur readm file file free text describ file manual ftp server directori structur read readm file order find sought resourc techniqu implement find resourc anonym ftp archiv site archi section analysi file ftp site present anonym ftp million textual document anonym ftp ftp site contribut wealth inform internet thousand ftp site total million roughli terabyt data internet distribut document deliveri system origin develop effort provid minnesota student staff flexibl inform system dissemin kind inform commun simpl singl interfac made popular rapid adopt site internet gopher evolv system primarili intend distribut document environ provid access type network servic gatewai initi popular gopher gopher space shrink gradual web document index retriev system gopher develop veronica present section file system transpar read access file distribut anonym ftp site internet access ftp file log remot host copi file alex ftp file part local file system file directori ftp site local file system root directori internet domain ftp site code alex structur simon fraser ftp site ftp berkelei file call local unix tool grep find find retriev document remot site make local copi track remot file updat document file put anonym ftp site alex inform index inform retriev servic web servic internet term data packet transfer nsf ftp servic internet ftp account larg amount data transfer internet hundr ftp site offer raw order find ftp site file archiv directori structur ftp download readm file eventu find file sought archi electron directori servic implement mcgill peter alan bill locat file archiv anonym ftp site automat catalog system period retriev list file anonym ftp site directori anonym ftp list combin searchabl access internet archi inform system gatewai archi file modif date ftp site path file found index produc month mirror site reduc traffic problem archi file index mean order locat portion find file archi descript file file topic archi extrem web system implement cnet provid index free softwar ftp web server organ topic veronica easi index internet gopher system simpl user interfac protocol simpl stateless protocol server request return document menu close connect make autom gopher space easi search document hierarchi menu eventu locat resourc develop make easi search item gopher space titl veronica period gopher space request menu gopher server start set regist menu index titl text menu receiv search term result menu link gopher menu search term menu entri veronica popular gopher gopher space veronica useless wai area inform develop machin system user search retriev document type resourc index wai develop collabor appl comput dow jone initi busi execut contrast archi index file wai index keyword content textual bibliograph descriptor graphic file archi veronica central index singl global index wai address scalabl approach decentr index wai divid index server provid inform server index inform locali regist repositori index directori oper machin directori index classifi server knowledg inform wai server server special resourc topic search user connect directori server select server search queri submit natur languag select server elimin conjunct remain word phrase appli full text index find relev document document rank base heurist word weigh algorithm relev document fed back server refin search relev success search automat run alert user inform wai system underli index scheme web search engin wai index inform retriev spider section show gener architectur wai approach distribut index unifi comput scienc technic report built marc indiana servic searchabl index exist technic document broadli relat comput scienc store anonym ftp site index pull inform larg number regist anonym ftp site format readm file regularli visit regist ftp site automat download readm file find readm file free text format merg larg master index technic report index ftp site file descript found readm file index technic report academ paper system advantag complet autom mainten index cumbersom readm file written free text readm file differ site site time ftp site ad regist ftp site index builder rewritten updat order account structur readm file site figur server directori wai server server select user queri figur wai gener architectur show gener architectur system internet directori servic develop network resourc discoveri project colorado fact locat document servic worth mention exemplari methodolog resourc discoveri attempt locat electron mail address inform internet user dynam develop protocol collect regist user internet difficult approach exist network servic locat machin sought user resid address machin finger servic strategi adopt intellig agent internet section regularli usenet archiv servic retriev uniqu address build internet hierarchi address futur search search queri submit provid login keyword describ institut locat user sought resid set heurist locat host desir user account mailbox queri refin user select locat guess subset domain select user search parallel advantag exist network servic success scalabl growth internet exist network servic resourc prove viabl prevail method suppli servic internet master index file http search engin index builder queri pool user www client ftp server cs tech report figur gener architectur interact multimedia inform deliveri servic web interact multimedia inform deliveri servic internet advent hypertext text retriev system research hypertext focuss user interfac author issu hypertext project wide area distribut global system access intern project cern led tim robert attempt make distribut inform easi physicist work world exchang data web born project rapidli adopt internet user interact multimedia support web reli hypertext document format html markup hypertext link document travel current document html annot document hypertext link intermix text imag media document web browser mosaic web wide interfac mosaic internet servic share inform internet simpl web user larg collect inform network log advanc inform web technolog choic deploi inform internet major reason tremend growth internet term amount publish inform term network figur show exponenti number web site internet gatewai internet servic develop web synonym internet huge amount data rapidli cumul web web find inform cumbersom find real inform process user drift end read possibl interest page irrelev origin phenomenon similar dictionari section highlight evolut inform retriev techniqu appli web catalogu directori number web page small start web list web server maintain cern suffic resourc larg number web server web page todai prohibit method inform retriev web start build list interest link made list publicli list ncsa meta index maintain list refer list user access refer resourc collect web rapid web site web page made list obsolet attempt make list automat collect refer introduc centr universitair dinformatiqu geneva base automat retriev refer fix set document report site automat sift group articl automat collect list refer resourc size rapidli larg valuabl resourc make list list made searchabl search kei user find entri refer list interest real index construct search process real time list text match string time dynam natur web made list stale impract catalogu directori built repres wealth resourc web rapid growth internet mainten list scalabl success catalogu web page yahoo initi graduat student stanford yahoo classif topic built hierarch tree node tree categori visual menu arc tree structur leav link web resourc resourc review classifi manual editor author submit url resourc web page web titl descript resourc yahoo editor submiss accept refer resourc classifi ad yahoo directori searchabl index yahoo directori implement contrari yahoo search engin web resourc larg yahoo searchabl index resourc catalogu yahoo editor index keyword resourc content index entri word yahoo categori resourc titl resourc descript submit author classif resourc user hierarchi find inform relat topic origin yahoo fact resourc necessarili individu web page autom discoveri index submiss describ entir web site accept content web page index user search directori topic submit term search case match resourc categori search term resourc titl descript search term document result rank prioriti base term final descript success yahoo copi concept present friendli user interfac review directori site looksmart approach present user initi list topic topic classif topic present origin list remain screen retain list continu select document list displai easili path reach resourc concept hierarchi classif looksmart hybrid search engin search catalogu combin result result search web space search engin popular type catalogu special directori refer resourc pertain topic directori refer web page relat directori creat manual author gener result queri submit search engin build search engin top special directori creat special search engin ring type special resourc directori circular list maintain web site pertain common central list site figur ring site topic collabor insert hypertext link web site link previou site circl site link site list site form ring central orchestr ring accept maintain server list link site ring user jump site ring randomli access site request link central list server figur show ring web site interconnect robot search engin simpl automat collect resourc solv problem inform retriev automat discoveri make effect wealth inform web space program download analyz store inform index queri document found explor graph hypertext link form document initi web link extract ad queue url resourc process repeatedli select url retriev page extract link program wander robot crawl web mirror statist calcul size internet estim todai number web site academ work describ robot work jonathan fletcher stirl implement work oliv mcbryan develop worldwid web worm search paper eichmann reveal anatomi spider built softwar engin program fund nasa wwww index text titl header html spider full text index hope improv recal system architectur system major parser index spider program visibl web download document found import note visibl spider limit connect graph document commenc start point start document web indirectli connect set start point ensur coverag directori web site good start point spider elabor compon search engin manipul larg url respect visibl web import note spider web page access dynam gener web page authent index receiv document download spider extract url link found document refer pass back spider download text analyz extract term index extract term differ system index full select term depend locat document analyz begin document compon index result index depend larg inform retriev technolog chosen system invert file vector space model probabilist index model call search engin fact program sift index answer user request interfac user index automat autonom explor web structur build searchabl index simpl eleg solut inform retriev web involv ethic concern relat result impact spider continu retriev entir object discard retain content robot gener substanti load web gener traffic internet spider index entir web content download plethora search engin entir corpu circul continu internet spider gener local load overload inform server continu consecut request document spider flood server prevent serv user precipit controversi base guidelin implement program web allevi load server suggest standard robot mechan spider part server part access spider follow recommend exclud superflu web page avoid server send consecut request server short period time research advoc techniqu index web interest spider creat index web page author write descript regist index summari process remov problem server overload web crawler gener pertin inform index submit author web site manag requir extra work page author end describ resourc notifi aliweb exist harvest system multipl layer approach support distribut index architectur harvest solv problem server load network traffic move index task inform provid site system major gather broker gather collect inform provid index spider broker interfac index origin solut varieti configur gather broker network broker special provid custom gather resid host inform execut period multipl layer approach built distribut repositori metadata describ artifact internet special similar essenc execut inform provid site site maintain time extract pertin data build local structur success distribut gener form structur layer gener concept held lower level architectur offer user possibl hierarchi layer interact pertin resourc structur approach advantag high recal technolog present multipl layer architectur knowledg discoveri potenti techniqu offer search engin work parser index index search engin lv spider page page figur search engin gener architectur figur show gener architectur search engin search engin spider index replic parallel crawl index index distribut central compon url list manag share local simplest list url visit start found url store list process page url remov store list visit page list avoid revisit page list page visit conform robot guidelin avoid revisit problemat site spider select url download page move url lv page download pass index index page extract store extract term page index entir web index index search engin answer user request index rapidli rebuild index restart web initi start point list start point list augment url submit user site index scenario simplist web execut web spider challeng task involv perform issu issu high parallel implement web due larg size web index web exist strategi url select spider store modif date page visit order revisit page updat regularli static page domain popular site revisit select url spider account fact consecut request page server recent page present techniqu url select base backlink count url referenc select visit spider relev rank document due commerci search competit push spider crawl index resourc order assert complet index search engin claim index million document complet import factor long search result adequ index internet portion wouldnt matter set match document set match document make big differ user user larg set identifi document fit quest demonstr singl web search engin user relev document resourc search queri yield match search tradit inform retriev rank resourc found relev present order relev size model chosen index user import differ search engin rank mechan relev subject term difficult high confid real user queri queri present context search engin learn past experi search engin analyz term relat term document higher relev document consid locat term html document inform locat word phrase term titl header consid relev word bodi document frequenc locat factor determin relev search engin stimuli influenc relev relev rank capit success search algorithm secur guard rare factor calcul relev document heurist larg support document higher frequenc keyword rank document low keyword frequenc frequenc calcul percentag keyword document order normal document size count frequenc word weight base locat document locat respect word titl gener consid import locat consid url anchor begin document word consid import paragraph close keyword factor word weight search engin disadvantag isol word favour keyword complet sentenc number match keyword queri import factor determin relev term queri match word document relev queri search term consid term document rank bf entireti creat cluster document rank frequenc search engin favour smaller document larger date modif document access easili date favour document instanc track modif date search engin updat frequenc document favour rank document updat regularli hybrid search engin maintain directori favour document catalog site catalog consid import web page referenc page import popular web page estim count number link lead page rank search engin favour page link point link weigh base popular web page notion link qualiti dublin metadata workshop stress import document network document facilit discoveri extens html specif includ tag descript keyword content summari insid html document figur show excerpt html document tag tag enter document author keyword reflect real content document relev rank web search engin favour page keyword survei search engin engin figur snippet html document meta tag meta tag match search term match term document bodi term extract meta tag keyword higher weight internet domain web page retriev plai role rank domain rank search engin compani pai page rank match page occur rank document import rare user dig deeper page document list statist show user page rank web page author web page rank commonli call spam spam consist ad keyword page keyword user comment tag colour text visibl web crawler common techniqu put keyword meta tag relat page actual gener keyword frequent keyword copi meta tag site rank high search engin techniqu heurist detect spam search engin penal detect rank lower techniqu rank retriev document recent direct hit direct hit technolog clever ibm almaden research lab pagerank stanford direct hit relev feedback approach user select usual search result record document select relev time queri submit technolog queri frequent direct hit gather inform user select queri inform direct hit rank proce direct hit exclud rank strategi supplement ad rank strategi idea complet metacrawl present section exploit user chosen refer metric search result rank metric account origin resourc select pagerank search engin stanford base basic compon rank metric link popular larg link graph call citat construct repres present link web page graph attribut pagerank weight web rank page search result mechan calcul pagerank page number link number page point pagerank dump factor pagerank repres probabl page random clever heavili base link frequenc rank document measur link main state simpli rank document search find authorit resourc pool web page object build directori yahoo directori automat find authorit entri categori start categori categori queri gener standard search engin result gather search engin constitut pool document order rank document authorit topic queri current directori categori page point page point document pool retriev add pool document page collect weight iter calcul weight link point carri weight page cumul weight iter page point import page weight clever page text add weight link search term text sort page authorit select repres catalog categori clever search result calcul weight time depend size root set classifi clever knowledg discoveri approach find network inform base relationship find resourc internet retriev al show due widespread synonymi polysemi natur index method base occurr singl word perform adequ method match word similar mean string pattern search tool find relat document search word queri search document retriev document concept sibl synonym broaden scope search includ thesauri thesauru correl term search engin conjoin search term relat term thesauri static base built dynam statist track word commonli queri document relev rank favour match search term match similar concept stem common techniqu retriev consist extract stem search term concaten suffix gener term gener poor case stem gener word stem advantag substr match document similar search consid type retriev word weight techniqu gener term weight vector repres document compar document similar function similar function present section start document similar search find document document term live document leap anim share similar vector similar research knowledg natur languag process pave automat techniqu gener queri lead content retriev approach concept hierarchi match concept concept hierarchi built expert synonym languag built automat subject extract assumpt agent inform retriev spider sort inform gather type agent special inform retriev agent base inform retriev stimul great interest past year definit agent question gener research relat field product heterogen knowledg cognit learn commun distribut queri field interest inform gather desir characterist agent allud research singl agent agent prototyp todai compound desir characterist agent includ commonli characterist distinguish agent form softwar compon agent initi control action modifi dialogu user clarifi request invok agent agent static chang sequenc action environ learn interact user agent custom prefer user agent commun engag involv commun user agent dialogu user disambigu request commun collabor agent user paramount inform gather task deleg agent perceiv plausibl behaviour emot state agent charact effect commun avatar repres act behalf user virtual space agent typic run host retriev inform scatter internet make resourc transmit real mobil agent agent transport duplic execut host exist agent mobil agent inform gather amount inform server greater transmit client mobil agent execut server inform gather filter mobil inform gather agent great import client machin disconnect lack process power resourc perform filter inform retriev agent program learn user adapt react collect retriev resourc knowledg satisfi network inform retriev agent person assist act behalf user reli tool servic user state agent determin find inform internet retriev simpl agent allevi burden resubmit request search engin automat send simultan user queri popular search engin agent specif featur search engin adapt search queri user rememb specif characterist peculiar queri interfac agent merg result elimin unnecessari duplic download web page result list scan content search term filter present relev result list similar resourc select search engin queri savvysearch learn identifi search engin queri track perform search engin agent search experi rank special search engin queri savvysearch adapt user adapt resourc approach similar inform agent matchmak present intellig facilit match inform provid consum agent adapt user observ user activ recommend resourc queri user system similar profil retriev submit agent manag user profil updat regularli base user interest activ web commerci base user behaviour predict user automat download local cach web page potenti interest user regularli access set page time dai stock agent automat retriev page minut usual access time cach local prepar user popular task agent inform filter inform filter internet context extract relev inform larg volum dynam gener usenet electron mail commonli dynam larg document collect inform filter internet dynam natur document repositori bind requir inform filter system continu run process sift incom messag filter messag satisfi user requir interest similar agent listen group channel alert user interest topic discuss agent summar articl interest appli approach web monitor found relev page agent report interest breed agent inform filter gather worldwid web shop agent access vendor web page bargain compar price offer suggest present user automat order product initi jango web shop compar product price offer creat account user initi inform gather agent time inform retriev domain domain return inform knowledg return resourc agent perceiv web mine process multimedia resourc discoveri imag video index retriev interest research field drew attent advent worldwid brought challeng comput vision artifici intellig commun tradit pattern recognit imag analysi algorithm vision ai field dealt small set imag scale larg collect imag video frame internet requir scalabl effici interest imag video retriev system begin web main famili imag video index retriev base content imag base descript imag imag retriev system suffer poor due term extract process automat assign keyword imag tricki task imag retriev system visual featur index imag system differ extract visual featur index queri system queri provid imag sampl system search similar imag compar featur vector extract sampl featur vector imag retriev system imag surfer provid base type search system queri sketch imag featur shape translat featur vector match featur vector qbic imag webseek provid queri imag featur specif queri combin imag imag number face imag content distinguish photograph figur visual featur extract limit imag retriev digit exploit retriev combin conjunct disjunct imag featur descript queri imag retriev system retriev imag imag surfer qbic larg imag repositori base object major differ imag retriev system domain index qbic index sole imag imag intern present web imag multimedia orient retriev index imag web index imag internet pinpoint web page imag graph web page imag easili discov resourc imag descriptor imag system give satisfactori result regard visual recal optim data mine knowledg discoveri internet data defin process extract unknown potenti inform data larg data mine princip core knowledg discoveri includ data data relev data pattern evalu knowledg visual data mine appli wide spread web technolog made larg document collect web ground knowledg discoveri research contrast resourc discoveri find retriev resourc knowledg discoveri internet aim deduc extract implicit knowledg necessarili resourc tradit knowledg discoveri function put time seri appli global inform network function attempt internet appli document repositori success text mine instanc attract interest import asset corpor organ collect document regularli technic web collect free text document embodi corpor cumul time essenc inform retriev system pinpoint relev document insuffici text drastic improv inform retriev system extract relev knowledg document allevi retriev document manual search pertin knowledg text mine includ knowledg discoveri data clean knowledg visual domin categori text mine text text document document visual keyword associ classif cluster visual similar search term categor document visual document text analysi text interpret figur text mine pyramid text analysi text document content investig syntact correl semant associ term kei concept phrase extract repres document document section text interpret abstract document form document content document categor organ document collect document visual consist document concentr group group intersect figur show pyramid text mine theme order categori pyramid illustr relationship text mine function function extract relev kei document extract limit term concept pertin topic document extract relev kei inform document essenti idea document abstract synopsi similar search document similar concept document organ collect document predefin theme set class descript document classifi document belong class search predomin theme collect document categor document found theme text mine part technolog constitut knowledg discoveri internet primarili act textual content internet highli dynam multimedia environ involv interconnect heterogen web structur mine web content mine web usag mine search result mine gener access pattern track custom usag track web page content mine web mine figur taxonomi web mine techniqu interact user text mine limit grasp knowledg environ data mine commonli call advantag content usag resourc relationship resourc web intersect data mine grow includ technolog found artifici inform field inform inform retriev document rank hyperlink featur usag categor web mine web mine defin topic continu fall realm figur show classif domain akin web mine web roughli knowledg discoveri domain pertain web web content web structur web usag mine web content mine process extract knowledg content document descript web document text resourc discoveri base concept index technolog fall categori web structur mine process infer knowledg web organ link refer refer web web usag web log process extract interest pattern web access log web content mine knowledg web buri insid document current technolog bare scratch surfac knowledg extract keyword web page result user search engin emerg human assist search internet web content mine automat process keyword extract content text document present approach suggest restructur document content exploit machin web structur effect web mine case intermediari reli built structur limit type set document typograph linguist properti natur document internet task technolog harvest reli structur document retriev inform usual approach exploit structur document wrapper map document data model declar languag queri data model reli rule repres web document graph tree extract knowledg restructur web document relat tabl advantag relat power data mine possibl techniqu lexicon content interpret group web content mine mine directli content document improv content search tool search engin web page summar research work retriev inform structur suggest approach limit group wrapper map content document intern promin research result knowledg discoveri heterogen irregular web present shopbot project special discov person homepag inform internet servic search engin retriev resourc relat person data heurist identifi typograph syntact featur insid document document person homepag weboql queri languag web page restructur graph tree web capabl retriev inform site cnn tourist guid shop agent learn recogn document site offer search servic human assist structur catalogu extract price list special offer agent capabl compil inform retriev site discov interest bargain search engin result summar heterogen web absenc structur lead research mine subset document data document pertain topic subset search result queri search engin system present small relat tabl minim inform provid queri languag result refin system access document retriev search engin collect inform document data provid server content content modif link declar languag abil retriev pertin document search result zamir etzioni present techniqu cluster document retriev set search engin techniqu reli sole inform provid search result snippet descript line page induc cluster categor retriev document discov cluster present repres view top list retriev document facilit sift larg search engin result list web structur mine interconnect hypertext web reveal inform inform document link point document popular link document rich varieti topic cover document compar bibliograph citat paper cite import method advantag inform convei link find pertin web page multi layer multi level internet level obtain success transform gener perform lower level lowest level primit inform store internet higher level store broad concept concept lower level save local specif relat tabl depend scope record describ characterist artifact found internet web page imag mean higher level cumul number artifact subsum concept hold power structur content global inform system web declar defin queri structur exploit knowledg convei inform network explicitli state document queri return list european provid web site document select affili document europ affili belong field public year count web usag mine web grow local server provid resourc simpl structur collect web access log web server record accumul data user interact request resourc receiv analyz web access log web site understand user behaviour web improv design collect resourc main tendenc web usag mine driven applic gener access pattern track custom usag track gener access pattern track analyz web log understand access pattern trend shed light structur group resourc provid web analysi tool exist limit unsatisfactori techniqu data mine onlin analyt process present appli data mine techniqu access log interest access pattern restructur site effici pinpoint effect target specif user specif sell ad custom usag track individu trend custom web site user inform depth site structur format resourc dynam custom user time base access pattern innov studi adapt web site improv learn user access encourag excit potenti applic web log file import success applic depend valid reliabl knowledg discov larg illinoi maintain list web access analyz hypernew page access raw log data current web server store limit inform access script site store addit inform effect web usag mine import clean data transform step analysi inform revolut publish sort document amaz speed overwhelm public techniqu manag excess resourc retriev pertin document develop growth term resourc term user access show urgenc crucial effici effect resourc discoveri techniqu survei outlin major inform retriev approach internet pinpoint weak major problem current technolog relev problem relev document search result make current search engin useless research effort underwai develop retriev web mine survei report initi undertak research necessit contribut collabor commun artifici data inform natur languag field born refer serg dallan jason jennif janet wiener lorel queri languag semistructur queri data int conf mark brian michael pazzani effect social discoveri filter web proc assist inform search av aho mj fast pattern aid bibliograph search commun june ai intellig system ai intellig petkov retriev visual review int multimedia tool novemb alex ftp filesystem craig knoblock queri process sim inform mediat laboratori plan schedul initi arocena alberto mendelzon restructur web proc icd februari jr al virag imag search open framework imag manag spie storag retriev imag video februari gordon bell jim prospect inform dream commun tim world wide web initi tim robert ari henrik arthur secret web commun august michael provid inform system full hypermedia function proc hawaii int conf system eric rob winslett secur access data internet bloom space time tradeoff hash code error commun mic peter michael schwartz customiz discoveri access system technic report mic peter michael scalabl internet resourc research problem approach commun august rs boyer js moor fast string search algorithm commun octob de bra post inform retriev make search feasibl jeffrei bradshaw softwar agent aaai press mit sergei brin lawrenc page anatomi hypertextu web search engin int conf april peter buneman semistructur tutori princip system confer bush atlant juli sr tomkin experi topic distil acm sigir workshop hypertext inform retriev andrea jai jerom yen intellig meet agent ieee august junghoo hector lawrenc page effici crawl url order int conf april georg track growth commun august jeff introduct survei ieee septemb bruce croft howard turtl retriev model incorpor hypertext link proc page novemb daniel rissland approach intellig inform retriev proc acm daniel rissland approach intellig inform retriev proc sigir page archi electron directori servic internet martin erik mark carol assess inform provid librari servic commun internet spring weld scalabl agent web proc autonom agent eh dl wp birmingham agent architectur michigan digit librari ieee softwar februari david eichmann spider balanc effect search web load proc www david terri integr structur system han erickson multicast backbon commun august oren etzioni gold commun oren etzioni move inform food deploi softbot world wide web ai oren neal richard segal build softbot novemb preliminari report oren etzioni daniel weld interfac internet commun oren etzioni daniel weld intellig agent christo faloutso access method text comput march christo faloutso dougla oard survei inform retriev filter method technic report august david yesha approach resourc discoveri journal internet fletcher internet robot structur front al queri imag video qbic system ieee page septemb imag search engin world wide web technic report cs univ imag search engin world wide web technic report comput scienc august jeff meta tag organ web page jeff web net norbert fuhr ulrich pfeifer inform retriev combin induct assumpt acm transact inform tk lm dumai vocabulari problem commun commun david john robertson hypermedia author ieee page andrew legion worldwid virtual comput commun david grossman frieder inform algorithm heurist kluwer academ jiawei fu resourc knowledg discoveri global inform multipl layer approach technic report simon fraser novemb jiawei fu resourc knowledg discoveri global inform scalabl multipl layer approach proc conf advanc digit hardi schwartz resourc discoveri system base semant file index proc usenix winter page rl processor text retriev septemb navig hypermedia system acm page adel daniel metasearch engin learn search engin queri ai michael munindar tomasz global inform manag local autonom agent proc int symposium gener comput workshop heterogen cooper knowledg michael huhn munindar singh read agent morgan kaufmann inktomi technolog hotbot white paper karen sparck jone peter willett read inform retriev morgan kaufmann literatur inform retriev hypertext septemb de jh vr pratt fast pattern match string siam june konopnicki shmueli system www queri proc ieee cs intl conf data koster aliweb archi index web proc int conf world wild koster guidelin robot writer nexor koster standard robot nexor koster web robot nexor koster world wide web spider robot nexor daniel kuokka larri matchmak inform agent proc joint conf daniel kuokka larri matchmak inform agent proc int joint conf page subramanian declar languag queri restructur web proc int workshop research issu data jane lee approach natur languag process phd harvard michael lesk ag inform retriev alon avi divesh maria challeng global inform system proc vldb illumin invari object model imag video retriev journal visual commun imag submit review bing yan imag retriev imag repositori chromat recognit kernel technic report simon fraser februari bing yan imag retriev digit librari illumin invari recognit kernel proc int workshop storag retriev issu imag multimedia int conf expert system august manber peter connect web search facil bulletin maron jl kuhn probabilist index inform retriev journal oliv tool tame web proc www kenneth corpor inform manag electron ag mendelzon milo formal model web queri acm symp principl alberto georg milo queri world wide web proc decemb merit data hara multimedia world wide web inform retriev engin proc sixth int www santa ted nelson file structur chang acm nation nguyen srinivasan access relat form world wide web proc acm sigmod conf manag june doug oard kim inform filter resourc daniel daniel robert plant artifici intellig virtual organ commun roger palmer onlin refer inform retriev librari rw sclaroff tool manipul imag spie storag retriev imag video volum page san perkowitz etzioni adapt automat learn user access pattern proc int world wide web santa april richard einer peterson biologi taxonomi smith data mine knowledg overview um advanc knowledg discoveri data page jeff crawl guid web pc riloff text inform retriev handbook comput scienc ab tucker crc rissland daniel cbr drive ir proc ijcai page august cs robert retriev method code proc decemb se robertson rank principl ir journal salton automat phrase match read automat languag process american elsevi publish compani salton automat inform retriev ieee comput salton bucklei improv retriev perform relev feedback journal american societi inform salton mj mcgill introduct modern inform retriev erik signal detect analysi www search engin design empir octob kyle net grown guid kyle search engin web directori guid michael alan brewster clifford neuman comparison internet resourc discoveri approach comput selberg etzioni search comparison metacrawl intern world wide web erik selberg oren etzioni search comparison metacrawl fourth intern world wide web page etzioni home page finder proc sixth world wide web santa april avi mike jeff ullman opportun centuri report nsf workshop futur system john smith stephen weiss overview hypertext commun juli jr smith sf chang visual search web content ieee mathemat analysi code method american februari danni search engin watch danni sullivan richard web search page dm sundai fast substr search algorithm commun august gari index internet elizabeth queri base technic report marc unifi comput scienc technic report lesson index resourc proc int www octob winslett web fourth intern world wide web page ronald chri charl ward mosaic web ieee octob peter vigil onlin retriev analysi strategi john lynn ward explor power internet gopher dec jan stuart jean eric miller workshop metadata essenti element network object march core stuart jean eric ron daniel metadata workshop report essenti element network object march core daniel weld oren etzioni law robot call proc aaai william william principl autom inform retriev busi wa wood import issu knowledg proc octob tak yan hector electron librari access worldwid inform ly jerri lee world wide web resourc discoveri system fourth intern world wide web page ct method detect describ gestalt cluster ieee tran januari andrew stephen veronica paul tarau retriev control natur languag proc workshop applic natur languag inform system june andrew stephen veronica paul tarau resourc discoveri natur languag search june zaian jiawei han resourc knowledg discoveri global inform preliminari design experi proc int conf knowledg discoveri data man jiawei han discov web access pattern trend appli olap data mine technolog web log proc advanc digit librari page santa april robert hobb hobb internet timelin oren zamir oren etzioni web document feasibl demonstr proc acm oren oren richard karp fast intuit cluster web document appendix glossari inform retriev web page attempt give definit term relev inform retriev glossari compil glossari origin scott weiss agent agent autonom piec softwar act behalf user task inform gather inform filter anchor hypertext start point link button click anchor link node anchor boolean queri queri boolean combin term exampl classif process decid categori document exampl decid newsgroup articl belong folder email messag direct gener topic essai cluster group similar document vector space perform retriev compar queri vector centroid cluster continu search cluster collabor filter process filter document determin document user similar interest found relev call collect group document user inform test collect collect fusion problem combin search result multipl collect tricki measur idf differ retriev fix number unclear collect filter process filter extract featur text document determin document relev call crawler spider document piec inform user retriev text web newsgroup sentenc book hyperlink web context link anchor web page commonli call hyperlink anchor index process convert collect form suitabl easi search retriev inform extract relat area attempt identifi semant structur specif type inform unrestrict text inform filter larg amount return data user standard problem ir inform user queri approxim inform inform retriev studi system recal text unstructur form document frequenc abbrevi measur term document collect defin document common word low idf word uniqu document high idf typic weight paramet model invert file collect essenti index word term invert file list document perform boolean queri link hypertext associ phrase imag node relat inform node link connect inform document navig hypertext process click anchor point leap node anchor point connect node hypertext context part document document broken smaller part call node node text scene video resourc normali convei idea theme standard measur ir defin number relev document retriev divid total number document retriev document relev widget collect system return widget xs ideal easi achiev return system attempt maxim recal simultan high retriev record meet exact inform term process compound term describ document page index term retriev term process singl term describ document combin base queri page index word glossari wed combin term base queri probabilist model model consid probabl term concept document satisfi inform bayesian infer net good framework style model inqueri system success queri string word inform user seek note english languag question queri process build queri creat ad term relev ad synonym term queri found question answer problem find exact answer user natur languag question larg collect recal standard measur ir recal defin number relev document retriev divid total number relev document collect document relev widget collect system return widget xs recal ideal recal trivial achiev retriev system attempt maxim recal simultan high recal retriev record german inform relev abstract measur document satisfi user inform system retriev relev document subject notion difficult quantifi relev feedback process refin result retriev queri user document return relev queri system typic find term common add queri return document queri repeat desir call similar resourc discoveri extract refer document relev queri robot spider rout similar inform problem retriev data continu stream incom inform signatur file collect document hash bit string essenti compress techniqu permit faster search similar measur alik document alik document queri vector space interpret close vector popular method comput cosin angl vector spider call program scan web url start web access link graph form web record inform server creation index search facil search engin creat spider problem written make larg number hit server short space system perform decai softbot softwar robot agent web search resourc servic tool collect refer stem process remov prefix suffix word document queri format term system intern model group word conceptu walk user doesnt specif queri porter stemmer algorithm task download sourc code algorithm dont rememb download stem word porter porter stemmer port hit document boat wine stopword word preposit articl semant content refer word high frequenc collect stopword term remov intern model document queri system predetermin list stopword stopword depend context word comput stopword collect comput scienc journal collect articl consum report term singl word concept occur model document queri refer word origin text term frequenc abbrevi number time term occur document queri count weight paramet model test collect collect specif creat evalu ir system set label human decid document relev queri tipster preval test collect collect classif reuter text categor test collect document articl label varieti topic design vector space model document queri convert vector featur vector word document stem remov stopword vector weight give emphasi term exemplifi retriev queri vector compar document vector closest queri consid return smart famou system vector space model web mine web mine data mine web includ techniqu aim discov knowledg web weight refer process give emphasi paramet import term vector space appli featur vector popular weight scheme scheme boolean term term frequenc vector weight normal sum divid squar root sum squar 
Agents/1030 1 oper schedul ea dept comput ny aarhu denmark abstract optim choic variat oper mutat crossov paramet perform evolutionari algorithm type oper gaussian remain entir run probabilist frequenc applic determin constant xed mutat rate recent studi shown optim usag variat oper ea run combin idea mutat oper schedul ea agent model spatial structur variabl size subpopul model select di erent oper type appli speci subpopul result optim choic oper problem vari handl approach oper schedul improv perform power superior compar classic advanc ea approach introduct evolutionari algorithm entail issu set popul mutat crossov variat oper previou studi shown set depend actual problem optim set chang run ea author tackl problem possibl ea paramet part genom evolv evolutionari process approach relat model ea individu mobil agent oper set prefer interpret spatial posit world oper paramet valu idea approach individu attract good paramet set track cluster good set optim set chang run problem suitabl variat oper type oper arithmet found deal problem paramet oper problem depend chang time uniform crossov valuabl begin evolutionari abil creat solut gradual diminish end ea run genet varianc left lawrenc davi introduc scheme deal problem oper idea adapt oper probabl base observ perform reward oper success closest approach techniqu present oper schedul variabl size subpopul approach structur popul subpopul type subpopul oper goal oper compet control shift subpopul success subpopul gain individu speci oper size subpopul determin global qualiti gain criterion updat gener subpopul model correspond classic island tag model structur insid subpopul previou introduc ea partli resembl idea di model world interact restrict neighbour model mobil island model motiv studi present paper superior perform compar simpl spatial ea techniqu possibl extend approach assign speci mutat oper subpopul call method oper schedul compar subpopul spatial structur insid relat subpopul size result local competit recruit individu di erent subpopul oper schedul ea interest human cultur religi concept histori impact human serv great sourc inspir popul model inspir line emerg ect religion popul religi rule typic includ commit convert provid mean propag religion member acquir reproduct religi educ strong competit di erent religion main inspir real world religion competit believ restrict mate individu belong religion result hybrid evolutionari algorithm spatial system compet subpopul consist popul individu world repres grid gure cell neighbour edg grid wrap form toru size world determin world dimens paramet refer cell individu religion de ne subpopul individu group individu grid cell movement direct interact rang fig world religion membership individu centr interact individu interact mate black individu top left convert grai individu religion vari run bound minimum believ maximum believ popul size remain constant individu consist attribut genom religion attribut de ne religion individu belong chang individu convert religion individu posit world determin locat genom de ne posit search space extend origin associ speci mutat oper individu mutat oper religion studi present number religion gaussian mutat xed gaussian mutat varianc anneal gaussian mutat varianc oper section algorithm shown gure procedur begin ea popul begin random individu mutat oper speci ed individu religion end end fig work individu evalu object function individu consecut perform action random evalu random walk action move individu randomli select neighbour provid individu chang religion believ convert individu convert neighbour di erent religion tness constrain individu convert religion minimum believ member left religion size convert individu larg maximum believ constraint avoid growth religion prematur mutat oper convert scheme simpler probabilist scheme origin turn superior preliminari experi execut mate action creat spring neighbour individu religion arithmet crossov spring parent substitut parent worst tness individu belong small religion size equal minimum believ mate individu religion neighbourhood mate action execut random walk convert action ensur spring replac parent parent opportun convert individu individu mutat mutat probabl pm mutat oper speci ed religion attribut tness score entir process repeat xed number tness evalu note indic individu transform avoid artefact sequenti applic convert random walk action mutat oper mutat oper studi gaussian mutat oper di erent varianc xed varianc scheme varianc anneal scheme varianc scheme introduc varianc xed pl power law distribut power law distribut number gener model shown simpli uniformli determin shape distribut studi xed experi order evalu perform experi commonli numer benchmark problem section short descript algorithm comparison simpl ea varianc mutat origin varianc mutat algorithm test instanc benchmark problem present section number tness evalu dimens experi repeat averag tness individu evolutionari process record fair comparison combin variat oper uniform arithmet gaussian mutat xed anneal varianc found arithmet crossov varianc mutat yield superior result compar oper choic benchmark problem report result combin oper set paramet popul size mutat probabl crossov probabl tournament select tournament size mention section gaussian mutat varianc arithmet crossov variat oper elit elit size solut found popul set popul size world dimens minimum believ believ mutat probabl paramet crossov oper part mate action gaussian mutat variant gaussian mutat oper section algorithm arithmet crossov elit size appli religion benchmark problem experi numer benchmark problem sphere step problem task global optimum result access qualiti oper schedul extens compar numer benchmark problem section tabl result experi column show tness found run tness evalu column show found solut run standard deviat small compar di erenc tness valu algorithm found solut benchmark problem compar converg speed turn slower begin run compar lower gene ow model restrict mate converg slower run inferior result indic problem perform robust nal result end run test algorithm perform compar figur show averag tness tabl result obtain experi numer benchmark problem problem sphere sphere step step curv individu algorithm benchmark illustr characterist case sphere step benchmark problem shown gure perform converg speed similar evolutionari process observ shift balanc power believ chang frequenc appli speci mutat oper gaussian mutat xed varianc favour begin mutat anneal varianc candid solut figur show exampl oper execut problem discuss paper introduc spatial oper schedul techniqu call extend origin ea algorithm hybrid ea spatial system compet subpopul subpopul speci mutat oper result experi show extens improv perform signi cantli capabl obtain good solut benchmark problem converg speed end run remark fast compar note algorithm superior compar simpl ea origin benchmark problem fig perform comparison graph show averag tness individu run versu number tness evalu iter anneal varianc iter anneal varianc varianc iter anneal varianc varianc fix varianc iter anneal varianc iter anneal varianc varianc iter anneal varianc varianc fix varianc fig religion size dynam religion process experi tness evalu problem obtain good solut low requir tness evalu interest real world applic tness evalu typic comput part algorithm good perform partial contribut spatial structur popul religion addit mate restrict reduc ow genet genet varianc popul high assign special oper religion advantag suitabl oper respect schedul oper control manner qualiti individu feedback mechan scheme success convert strategi investig preliminari experi interact rang world plai kei role converg abil investig paramet lead addit perform improv result optim choic oper vari run problem depend futur studi plan investig method advantag real world multipl sequenc align evolut hidden markov model topolog problem bene oper evolutionari process interest compar oper schedul compet subpopul scheme oper reward strategi davi acknowledg author colleagu valuabl comment earli version manuscript work support danish natur scienc research council refer paramet control agent base model congress evolutionari comput vol san usa genet algorithm pb european confer arti cial life mit press compet crossov adapt ga framework ieee intern confer evolutionari comput je adapt mutat rate steadi state genet algorithm ieee intern confer evolutionari ieee press adapt oper probabl genet algorithm jd intern confer genet algorithm strategi adapt compet subpopul intern confer parallel problem solv natur spatial model evolutionari algorithm hp parallel problem solv natur cellular genet algorithm intern confer genet morgan kaufman al handbook evolutionari comput iop publish oxford press appli critic evolutionari algorithm hp parallel problem solv natur introduc spatial model critic evolutionari algorithm master denmark 
Agents/1863 1 busi process agent john comput scienc sydnei po box broadwai australia abstract architectur process agent architectur process agent intend deal corpor sensit corpor rule agent adapt behaviour basi likelihood plan estim cost plan introduct intellig system societi autonom cooper compon maintain ongo interact environ intellig agent cooper adapt process agent architectur design specif busi process applic typic cost bring busi process substanti due cost human process involv process manag system make consid base perform process agent architectur improv estim likelihood plan estim cost plan term wide rang mean research literatur term agent comput situat capabl flexibl autonom action order meet design term system refer type system multipl autonom busi process manag establish applic area agent technolog term busi process refer process routin natur reli level initi system bring chosen busi process place demand process manag system process agent architectur trial local busi process applic process agent architectur varieti architectur autonom agent fundament distinct intellig agent architectur extent architectur exhibit delib reason reactiv reason agent architectur combin form reason call hybrid architectur report class hybrid architectur bdi agent architectur member class interrap architectur origin work process agent architectur extend interrap base set basic concept concept belief social world goal cooper local goal trigger cooper local trigger procedur plan intent cooper local intent procedur process agent architectur consist conceptu architectur control architectur conceptu architectur describ agent control architectur describ agent oper process agent architectur differ architectur interrap architectur appli design load dock case studi applic safeti busi process applic system thing robust agent architectur suit circumst gener busi process agent architectur intend specif busi process applic process agent interact user world wide web role process manag system direct commun user manag clear manag storag retriev send email messag user process agent commun user modifi user trai send user email messag user commun agent place document user achiev softwar document document user trai agent incom agent belief addit direct commun user busi process system manag perform automat document check process agent commun agent send messag user process instanc uniqu identifi extern transact creat process instanc final number agent belief relat instanc remain garbag collect process remov belief basi commun protocol gener busi process agent sender messag life messag deleg kqml queri commun process agent messag area agent agent post messag agent bs messag area agent messag manag role agent messag area messag instruct messag post messag remov sender receiv agent messag manag displai messag receiv agent messag area state condit remov sender receiv agent messag manag remov exist messag receiv agent messag area messag remov condit express term achiev goal trigger messag express term receipt messag origin messag agent ur conceptu architectur gener busi process agent shown figur architectur world belief deriv read messag receiv read document involv process read messag receiv agent activ fundament document read inform requir user agent send messag feel social belief agent observ agent system belief plai role partli wholli activ local cooper trigger lead agent commit initi intent plan achiev messag call delib reason partli wholli activ trigger pass data partli execut procedur advanc agent progress achiev goal commit messag opinion lead belief activ procedur call reactiv reason reactiv reason involv procedur procedur trigger watch arriv high prioriti messag social world belief agent user doc coop local proc goal coop local proc intent procedur trigger import data belief messag area messag manag read figur conceptu architectur delib reason hybrid architectur process agent architectur exhibit delib reactiv reason delib reason manag framework framework deal natur plan failur plan built exist repres failur abort power approach insuffici busi process plan execut successfulli necessarili relat plan goal achiev reason inevit fair amount activ busi process process manag system process agent emploi form elabor plan busi process necessarili reli achiev goal chosen path plan achiev plan fail execut plan goal achiev activ extern busi process manag system plan bodi call success condit procedur goal determin plan goal achiev success condit final path plan success condit execut procedur succe fail abort execut success condit succe success plan unknown plan exit result attempt execut plan shown figur sc plan goal figur plan exit plan bodi direct node label plan bodi usual condit construct iter construct gener plan block shown figur diagram plan bodi start state condit activ action stop state label success state action fail state action unknown state action abort state condit abort action success state mean plan permit execut freeli plan goal achiev fail state mean plan permit execut freeli plan goal achiev unknown state mean plan permit execut freeli plan goal achiev abort state mean plan permit execut freeli attempt build plan block level plan sixteen exit plan shown exit exit denot abort unknown state fail state de rea busi process model event process chain primit model control logic split unit defin inform object data busi process model fine level develop structur process agent plan plan plan goal start plan bodi figur gener plan busi process manag system compar system oper level polit natur human distinct made corpor corpor rule statement thing work cultur context knowledg make system knowledg thing made corpor rule state member work dai cultur context suggest receiv copi hour rest fred bit case system fred system work user role process manag system control process instanc user plan shown figur mean distribut individu set rule plan defin simpli term long oper constraint violat plan figur meet held short notic cultur type illustr figur invit receiv copi hour rest conflict form avoid attach prioriti cultur fred cultur context model separ rule cultur context describ make system work rule describ function chunk cultur appli wide rang fred receiv sort materi colleagu rule cultur repres separ model process cultur repres goal plan modifi rule long oper constraint violat start sc figur cultur adapt delib reason mechan bdi agent typic involv basi observ commit select plan select plan consist set thing agent appli procedur requir mechan identifi select plan schedul intent problem select plan set plan path composit plan disjunct problem call plan select plan perform badli plan perform defin term rate plan cost time execut ad busi process combin measur uninform agent made perform perform uninform agent expect make long agent perform stabl mechan achiev form adapt reinforc learn plan uncertainti discuss detail altern approach base probabl approach base principl probabl option probabl option principl appli valu discret continu variabl inform agent thing idea reason option lead failur reason translat probabl combin estim give inform estim agent perform account reason agent behav rmed adapt agent perform measur support adapt aim agent adapt make good choic execut avail plan goal flexibl choic plan goal presenc disjunct node plan flexibl choic path plan perform path plan measur avoid unnecessari descript refer measur method appli equal plan success execut path success plan success condit satisfi plan succe avoid wast time resourc annoi busi process system mean path relat likelihood path lead cost achiev time achiev path ad process achiev path paramet estim probabl path succe time execut expect time path execut expect cost execut path expect ad process execut path event previou termin path order recent termin termin path measur made observ time path execut observ cost path execut observ ad paramet binomi approxim distribut standard condit paramet assum distribut shape distribut paramet depend mean distribut assum normal li rang point estim extent estim emphasi recent behaviour determin chosen reflect expect pattern behaviour path estim geometr iter updat valu occas path execut point estim commenc iter initi init repres optimist start point assumpt hold requir estim point estim standard deviat distribut estim geometr fa estim recent observ behaviour path commenc iter initi init repres optimist start point method updat fa valu requir point estim statist path choic contribut virtu formula updat valu valu express term behaviour factor twenti trial contribut weight ten trial contribut weight trial contribut weight method estim probabl distribut path plan lead success cost path plan estim measur real nonneg argument similar estim estim mean standard deviat distribut paramet perform measur support choic path plan shown figur paramet likelihood success time cost figur perform estim process agent adapt adjust behaviour basi observ perform chosen path probabl path select agent probabl path path mean path plan term likelihood path time path path path probabl path path probabl estim probabl area normal distribut standard deviat mean method estim probabl path method extend estim probabl path choic number path info rmed adapt method path select select path estim probabl path path estim base observ past perform method account happen path select account belief present environ effect futur perform belief infer past perform expect effect futur perform inform agent address issu inform agent belief present busi unavail belief treat hypothet evid modifi uninform estim futur perform likelihood time cost uninform estim modifi give inform agent belief past belief infer past perform inform agent reason past select lead failur agent believ reason effect futur reason quantifi tempor effect belief translat estim likelihood cost belief lead estim reason select path plan behav extract observ interact user agent involv path present process agent make observ attend instanc process yesterdai attend instanc process schedul complet infer belief give observ perform present attempt made gaug rel belief uninform estim process agent make conserv tempor bound uninform estim process refin form simpl factor uninform likelihood success process agent implement java implement interpret agent specif interpret enabl agent built quickli simplifi deal high level specif featur process agent architectur method delib reason account polit corpor cultur corpor rule method agent adapt account happen path select account belief present environ expect effect futur perform belief infer past perform expect effect futur perform refer jk workflow system intern confer practic applic intellig agent march jk architectur support make eighteenth intern confer knowledg base system appli artifici applic innov expert system vi cambridg decemb labrou mayfield kqml agent commun languag jeff bradshaw softwar agent mit mn huhn mp singh manag heterogen transact workflow cooper agent nr jen agent applic market pp nr jen mj wooldridg agent applic market nr mj wooldridg roadmap agent research develop autonom agent kluwer academ jp mller design intellig layer approach note comput springer tj nr faratin eh mamdani design implement architectur busi process manag jp mj wooldridg nr jen intellig agent iii rao mp georgeff bdi theori practic int conf system san pp june rs sutton ag barto reinforc learn mit weiss system mit mp wellman formul tradeoff plan uncertainti morgan kaufman busi process model workflow integr approach stickel busi process model 
ML/2580 6 genet algorithm georg harik fernando lobo illig report januari illinoi genet algorithm laboratori illinoi transport build mathew avenu il genet algorithm georg harik high school mountain ca fernando lobo dept engin universidad nova de lisboa mont da portug januari abstract user point set paramet genet algorithm trivial task user typic interest popul crossov select ga technic interest solv problem blackbox simpli press start button paper explor develop ga fulfil requir paramet whatsoev develop algorithm account aspect theori includ previou research work popul schema build block genet drift introduct holland genet method easi applic wide rang problem thing stand current genet algorithm requir bit order make work applic user gener decid code decid good set paramet valu ga robust genet algorithm turn method lot tune paramet state affair kind contradict holland origin goal user make appli ga group categori choic code oper choic paramet set paper address categori unimport irrelev import topic subject extens research motiv work linkag learn reliev user special code oper linkag learn ga detect import similar oper effici process build block work similar flavor categori choic algorithm reliev user set paramet ga visit student illinoi genet algorithm laboratori result ga make life easier step closer direct algorithm simpl remain paper organ section review work relat topic paramet set ga section describ genet core paper section present comput simul show ga action paper end suggest coupl extens work relat work section review import research effort understand relationship affect attempt elimin ga substanti amount research paper briefli mention small part divid work empir theoret paramet adapt empir studi systemat studi understand relationship ga conduct de jong combin paramet set function base dejong conclud test set paramet gave good popul size crossov probabl mutat probabl dejong publish paramet set adopt research refer set dejong work gave practic guidelin subsequ ga applic research downsid peopl result blindli appli solv kind problem theoret studi dynam ga hard studi effect paramet simultan research studi analyz effect paramet studi proven insight ga dynam relev work select mutat popul size control map work popul size special result model show set popul size rang problem huge mistak work control map kei develop ga control map region paramet space ga expect work talk issu paper paramet adapt paramet adapt techniqu chang adapt paramet valu search progress topic investig earli dai evolutionari comput field techniqu includ central decentr control method central method chang paramet valu base central learn rule smith decentr method central class algorithm call evolut strategi idea decentr method encod paramet crossov mutat individu string paramet subject rule evolut approach paramet adapt work idea run higher level ga search good set paramet lower level ga higher level run popul individu encod paramet lower level ga regular ga set found ga work paramet adapt adapt mutat crossov work adapt popul size scheme work smith except address detail section develop ga previou explain motiv remov paramet genet algorithm readi present ga work ignor mutat time doesnt mutat unimport extens debat evolutionari comput commun crossov versu mutat recogn mutat shouldnt discard lightli work ignor sake simplic revisit topic mutat end paper satisfi ga paramet affect perform popul select crossov probabl remain explain elimin paramet exposit proce step rid select rate crossov probabl rid popul size rid select rate crossov probabl select rate crossov probabl elimin singl shot proper understand role paramet select rate user control amount bia individu crossov probabl user control amount recombin mix paramet work ensur build block grow gener gener closer schema theorem reveal surviv probabl schema made select rate crossov probabl previou research shown ga robust set paramet observ build block compact ga work wide rang combin import thing respect schema fall extrem case low high select pressur low select ga discrimin good bad individu high select ga pai attent result loss recombin extrem case easili exclud set select rate build block compact gene locat close chromosom string low high ga obei schema make growth ratio build block greater effect select oper schema gener disrupt factor schema due crossov oper net growth ratio schema gener express simplif schema theorem conserv hypothesi schema destroi crossov substitut formula obtain growth ratio schema set net growth factor ensur build block grow mix singl individu matter adequ popul size build block popul size requir reason ga learn linkag popul size requir extrem larg rid select rate crossov probabl set problem rid popul size popul size critic paramet ga ga converg solut ga spend unnecessari comput resourc theoret model address topic popul size ga essenc model recognit ga make mistak decid build block competitor size ga sampl build reduc error simpl model dictat popul size larg ga correctli statist build block competitor result equat popul size proport problem build ratio model difficult appli practic reli assumpt hold problem order appli popul size user maximum level decept select advantag build block tough competitor inform hard estim equat accur proper mix build block due difficult appli equat practic easi appli process autom algorithm smith attempt order author requir user desir accuraci level select loss build block signal goldberg al suggest algorithm onlin estim schema fit size popul select loss approxim user target loss result algorithm limit hard relat user select target loss actual accuraci ga solv problem estim schema fit varianc ga sampl partit simultan popul size theori assum build block mix occur practic work smith studi attempt automat adapt popul size difficulti exist order appli popul size theori practic step back observ user typic solv problem ga start small popul size larger popul size feel problem respond popul set situat familiar ga user includ user kind singl algorithm present approach rid popul size major drawback logic consequ constitut effect mechan elimin popul size paramet doubl popul size converg approach consist iter run ga popul size start small popul ga run converg ga fire larg previou converg process repeat ad time popul popul size doubl ga work forev stop comput run user happi solut press stop button power roughli mimic user typic ga systemat doubl popul size time popul time ga order reach solut factor time ga start optim popul size user doesnt wait press stop button earli ga ga optim popul size techniqu drawback immedi obviou occurr genet drift lead long converg time genet drift occur pressur discrimin distinct solut simplest illustr problem individu individu fit popul half individu half individu ga run select pressur discrimin popul converg process random wait time converg long roam popul call drift process occur ga consequ wait time popul size doubl long unaccept solut avoid long wait time detect popul affirm fire popul converg drift detect difficult task stop popul run fit varianc popul member fall threshold approach introduc problem introduct philosophi work problem noisi fit function threshold ffl converg mean individu genotyp notic eventu mutat turn reach drift addit paramet establish race multipl popul altern approach tackl drift consist run multipl popul simultan idea establish race popul size ga advantag smaller popul give function evalu smaller popul chanc converg faster larger smaller popul head start start drift catch larger popul smaller popul destroi point larger popul averag fit greater smaller ga rid smaller popul larger popul higher fit smaller point continu run smaller smaller produc fitter individu larger drift problem partial elimin popul start catch larger popul coordin arrai popul easili implement counter illustr tabl tabl mechan ga counter action base run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul run gener popul time counter base posit digit chang increment oper note posit popul ran algorithm initi run popul run popul popul popul illustr tabl popul run time gener popul descript popul size previou popul account number gener popul observ popul spend number function evalu popul popul converg popul point algorithm remov reset counter averag fit popul averag fit larger smaller popul counter reset typic state ga shown tabl ga run popul size smaller size gener popul size gener larger size gener tabl typic state ga popul popul current averag index size gener fit result section present comput simul test compar perform ga regular ga paramet set problem ga implement simpl ga simpl ga fix code oper simpl ga implement tournament select uniform mutat independ run perform order result statist order compar ga regular run algorithm target solut record number fit function evalu test problem carefulli chosen illustr specif aspect ga problem count problem easi problem exist popul size theori order compar ga ga popul size noisi problem problem illustr multipl popul run simultan order overcom drift problem bound decept problem user knowledg locat decept build block simpl ga unabl effici mix build exist popul size theori hold function simpli return number individu string test string length bit target solut string paramet regular ga tournament size probabl crossov equal popul size paramet chosen optim problem ga popul size chosen base popul size equat shown bb ln ff popul size depend build block size number build block build block fit varianc oe bb build block fit signal bit valu oe bb plug valu ln ff ff probabl ga make mistak build block ga expect correctli solv proport ff build block ga optim solut run ff impli infinit popul size practic popul size correspond small error probabl figur compar regular ga ga regular ga averag function evalu order reach optim ga averag function evalu ga optim paramet set reach optim solut faster ga ga bad knowledg ga reach optim solut time time optim ga popul size scheme factor rest due select rate crossov probabl ga easi problem higher crossov rate aggress select achiev higher solut qualiti faster price paid order robust set paramet set work easi difficult problem notic clever trick count number function evalu simpli count product popul size number gener care ga spend half number simpli popul member dont undergo crossov function evalu figur problem line solut found algorithm averag run solid line regular ga dash line ga noisi illustr multipl popul run simultan ga test noisi problem fit individu fit problem absenc compon practic compon simul toss gaussian random variabl varianc oe string varianc oe correspond larg amount miller studi effect popul size requir extend harik al equat account result equat oe oe ln ff fit varianc oe level oe obtain popul size ln ff level accuraci popul size figur show comparison regular ga optim paramet set ga regular ga averag function evalu order discov individu build block function evalu figur noisi problem line solut found algorithm averag run solid line regular ga dash line ga ga averag function factor trace observ smaller popul ran converg catch larger popul due larg amount tini select popul simpli drift notic drift complet elimin popul dont ran converg bound decept test case illustr problem difficult simpl ga problem concaten copi trap function trap function function unit number defin fit function sum independ previou knowledg locat build hard simpl ga combin build block individu assembl singl individu order simpl ga obei schema requir larg popul size exist popul size model unabl predict accur popul assumpt build block mix hold desir target solut string build block solv correctli order obei schema regular ga select rate crossov probabl minimum popul size order solv run optim found empir figur compar function evalu figur copi trap function line solut found algorithm averag run solid line regular ga dash line ga ga regular ga averag million function evalu reach target ga number extens coupl extens top work integr mutat ga outperform ga problem benefit combin question combin effici empir theoret evid ga mutat work small popul crossov work larg popul size ga ga total strategi question integr mutat crossov difficult area addit research extens experi kind ga import mention ga ti specif implement valid kind ga simpl restrict simpl ga order make exposit clear conduct experi linkag learn ga result power algorithm reliev user set ga reliev user special code oper result report summari paper present practic approach elimin paramet ga illustr oper ga implement ga crossov probabl select rate order obei schema fall extrem case high low select pressur fix select rate crossov probabl approach similar adopt paramet set paramet set back sound theoret constitut robust set obtain perform tweak interest peak perform interest robust simplic popul ga tackl run multipl popul rid drift ga bit slower ga start paramet truth paramet set arbitrari problem techniqu present requir user press start button algorithm run user happi solut wait long decid press stop button user ga technolog dont ga specialist order dissemin usag algorithm simpl issu paramet set user ga remov great part make life easier strongli contribut success applic ga theori ga base solid narrow exist gap ga theori ga practic refer optim mutat rate genet search intern confer genet algorithm san morgan kaufmann evolut strategi variant comput implement genet algorithm engin comput scienc pp john wilei son adapt search simul evolut unpublish doctor ann mi adapt oper probabl genet algorithm intern confer genet algorithm san morgan kaufmann de analysi behavior class genet adapt system doctor ann arbor compar analysi select scheme genet algorithm foundat genet report genet size popul complex understand mix genet algorithm journal societi instrument control optim control paramet genet algorithm ieee transact volum ieee learn gene linkag effici solv problem bound difficulti genet algorithm doctor ann arbor illig report genet size popul fourth intern confer evolutionari comput ieee press adapt natur artifici system ann michigan press adapt search reproduct effici genet algorithm doctor urbana illig report genet algorithm hillclimb parallel problem solv ppsn ii adapt resiz result complex mix genet algorithm intern confer genet algorithm san morgan kaufmann 
DB/2622 3 wrap queri languag effici xml integr vassili christophid sophi cluet jerom simeon institut comput scienc inria rocquencourt bell laboratori po box bp mountain avenu greec le chesnai franc murrai usa abstract modern applic digit requir integr access inform sourc tradit dbm semistructur web fast low mainten cost rapidli evolv environ interest xml middlewar model applic xml enabl fast wrap declar integr queri process integr system penal lack algebra adequ optim properti difficulti understand sourc queri capabl algebra approach support effici xml queri evalu defin gener algebra suitabl semistructur xml queri languag show algebra type wrap structur queri languag oql sql develop optim techniqu integr system introduct xml wide develop web applic requir data integr fashion account xml justifi technic ground xml enabl easi wrap extern sourc declar fast cheap mainten applic system effici tradit integr softwar address issu motiv xml technolog improv plan build web site provid project support iv project applic program claud monet impressionist claud monet waterloo bridg impressionist paint oil canva claud monet doctor figur sampl xml data cultur good access commerci inform cultur good integr highli object trade partial structur document repositori support descript inform artist work figur show sampl xml data export sourc advantag build applic xml due flexibl data xml repres structur semistructur inform figur easi convert data gener fashion independ sourc languag support declar integr xml data msl struql ya xml facilit queri process integr system hard issu ffl wrap type inform reason preserv type inform import queri optim data manag system export data dont provid type inform xml current form type dtd suffici captur rich type system object partial structur document figur work mandatori element element histori recent xml schema studi definit standard introduc type suitabl repres mix valid xml rest paper ffl wrap sourc queri capabl internet sourc export data provid queri facil order integr understand import perform push process sourc applic avoid massiv data transfer reduc xml overhead techniqu tsimmi queri templat describ sourc capabl exhaust descript sourc capabl find queri feasibl templat impli costli ad hoc order wrap set queri applic ffl process xml queri effici integr context remain open problem algebra support peculiar xml languag miss abil exploit partial type inform heterogen sourc capabl algebra framework optim techniqu address issu make algebra xml introduc oper model base algebra xml algebra express captur semant exist structur queri languag sourc descript languag show algebra wrap full text queri structur queri languag oql sql complet queri languag set gener effort requir applic queri process techniqu show algebra optim integr applic introduc rewrit techniqu queri investig impact type inform queri process illustr queri evalu advantag sourc queri capabl cultur art run run run connect connect import import load figur instal wrapper mediat paper organ section illustr advantag xml integr explain step requir build applic ya integr system section recal specif type system section introduc algebra descript languag wrap sourc queri languag present section present optim techniqu section conclud section xml integr ya system semistructur data system turn xml integr system reli librari gener wrapper declar integr languag call ya figur illustr step requir setup applic ya ffl simeon wrap object simpli run program export structur inform art system queri capabl wrap section ffl wrap cultur sourc gener wrapper wrapper understand xml type type system fulltext index wai expect paramet standard wai sourc configur file ffl cluet run connect wrapper port number fellow import structur queri capabl connect system load favorit integr program odmg model work work model symbol string field field schema set artifact class tupl artifact int string titl string creator year float price owner list set class tupl string string entiti auction structur work string string string titl style string artist work field size class class type symbol type type tupl type symbol set bag list arrai bool float figur mediat structur metadata closer integr give structur inform export wrapper note wrapper mediat commun structur oper xml type system repres structur inform level gener relationship level captur instanti mechan recal briefli figur show graphic ya model type inform import wrapper left figur repres data conform odmg standard schema art note bold font denot pattern symbol denot symbol denot multipl occurr altern type atom collect refer class pattern tupl pattern turn collect associ attribut class artifact concret instanti tupl attribut lower part figur repres inform export wrapper document mandatori inform possibl addit field illustr abil ya type system captur partial structur inform top part figur show ya captur pattern import properti artifact structur recogn instanc model artifact odmg section queri languag wrap advantag mechan integr program declar languag sequenc rule queri partial result connect skolem function give ya queri integr program queri construct collect document combin inform sourc make doc work owner match artifact set tupl list tupl work work queri consist match perform filter navig structur data bind variabl inform interest artifact titl variabl list option xml element ya filter mechan reli tree instanc deduc map node valu variabl type error occur note unambigu filter involv unambigu regular polynomi time fulfil usual function make construct result creat pattern valu return previou build tree distinct artifact group doc node skolem creat tree identifi distinct valu titl creator skolem function identifi creat refer note type inform provid wrapper ya program guid integr check applic consist notifi integr administr sourc modif technic challeng queri process illustr simplic integr qualiti structur descript provid ya system tsimmi mix offer similar function evalu user queri effici invit proce assum notic creation place issu artifact creat make match order process address view definit access semistructur field understand sourc answer queri exploit wai textual queri avoid download document ya oper model algebra choic oper model remaind descript sourc capabl queri optim support express power captur evalu exist featur notabl complex pattern match primit order navig xql object creation support flexibl type xml favor flexibl xml queri languag type wrap structur languag oper model support flexibl type filter strict form type oql support optim algebra equip number offer interest optim opportun oper model base function fix set predefin function ya xml algebra model function note skolem function algebra inspir object algebra present newli introduc requir deal tree briefli recal show queri translat oper model give overview altern algebra ya xml algebra main characterist xml data arbitrarili nest adopt techniqu similar algebra start arbitrari xml appli call extract relev inform produc call compar relat tab appli classic oper call gener nest xml structur bind oper extract data input tree filter tree distinct produc tabl variabl bind result figur bind oper appli tree repres xml collect filter bind work titl artist style size option element variabl collect note similar tab structur relat bind oper support type vertic horizont navig regular express section rewrit bind simpl oper tree oper appli tab structur return collect tree conform input pattern figur tree oper appli result previou bind denot work group artist group primit titl work tree bind work work work titl artist style size field si field si histori techniqu tree oper bind oper figur bind tree oper skolem function creat identifi perform assign previou creat identifi artist skolem function creat valu side effect integr view orthogon rest algebra oper object algebra intersect relat classic object oper map depend navig nest collect definit tab structur collect tupl straightforward recal definit explain oper appli top level tab structur manner similar relat extra bind appli algebra standard advantag optim properti rewrit techniqu object context relat nest queri remark bind tree frontier oper isol process standard call algebra case captur gener path express optim address import aspect algebra independ underli physic access structur reason evalu xml xml data local store document manag system xml virtual access wrapper section present rewrit case ya algebra translat figur show algebra translat ya view definit present section queri xml queri languag perform similar manner obtain translat document input oper algebra express match statement translat bind oper captur creat tab structur process predic involv input translat join oper predic translat select oper make translat tree oper relat work lore algebra physic aim optim involv index sal logic xml provid express power algebra logic suffici express relationship algebra unclear provid simpler version bind oper term regular express complex bind serv exploit sourc capabl compar object bind resembl scan oper potenti complex object algebra oper similar skolem function present note translat structur preserv involv queri tree bind select work titl bind work work titl artist style work tree join view definit titl year price owner style work doc artist size size doc select bind artifact set class tupl class tupl auction person owner list artifact year creator price titl figur algebra ya queri wrap queri capabl explain section wrapper export sourc capabl explain inform commun mediat show combin oper model type system level full queri languag oql odmg set queri method textual predic xml wrap sourc oper ya perform step concern signatur semant step access oper assum schema featur specif current price class artifact import wrapper xml input output element current price artifact return float declar perform automat wrapper schema manag method made mediat step requir optim wrapper perform step automat sourc featur oper captur core oper step manual issu discuss section explain captur oql wai describ oql capabl descript oql sql similar manner wrapper implement complex due natur captur bind capabl ya oper model borrow larg part oql algebra ya captur opposit true oql bind capabl restrict queri schema order restrict distinguish bind oper evalu oql understand accept filter oql figur show specif valid filter call xml serial type pattern figur possibl annot flag bind flag correspond filter restrict bind flag node tree label variabl inst flag label edg complet instanti left unchang filter class line actual object valu bound line extract class schema inform prevent figur filter pattern oper interfac line class schema specif filter instanti line oql oper figur show larg part oper interfac export wrapper oper kind declar algebra oper bind signatur special export pattern algebra oper evalu follow special signatur oper appli tab structur result collect tupl valid odmg data note oper push data import sourc result push oper argument oper select push predic function method current understood case integr bind select oper side figur push translat wrapper oql select describ wai capabl basic oper entri point seemingli simpl oper support web site search access queri interfac export full content capit understand oper support captur origin ya algebra appar straightforward assumpt retriev queri true protocol wai retriev engin wide digit base textual queri protocol establish clear separ retriev queri artist style element export xml document queri option field extens oper declar predic queri field export mediat import queri capabl sourc show wrap fulltext capabl sourc declar sourc declar support bind select declar fulltext predic suppli wai give part note pattern permit bind full document work achiev mediat semant push sourc connect exist equal predic exist algebra queri work impressionist artist evalu search work string standard evalu equal predic mediat express give readabl form origin state start select equal result bind denot arbitrari variabl add gener predic root document relat work garlic sourc capabl code programm wrapper remain unknown commun wrapper time part queri accept remain process disco descript sourc oper entail extra work optim order match gener plan import queri descript tsimmi optim opportun reduc interfac languag capabl describ set queri full queri languag ya system gener complet descript queri capabl structur sourc heterogen environ optim queri capabl point earlier optim techniqu relat object appli directli oper algebra introduc rewrit techniqu bind tree oper xml queri bind rewrit bind oper captur power featur xml queri vertic horizont type filter potenti crucial understand simplifi rewrit bind simpler bind chanc push sourc bind entail navig costli transform tradit associ access bind vertic navig upper left part figur show bind oper algebra translat view definit bind correspond vertic navig set artifact local attribut inform set owner navig nest collect captur object algebra join input depend left algebra bind shown upper middl part figur bind split elementari connect appli classic rewrit transform navig associ upper part figur exploit person extent transform standard join support effici evalu algorithm complex bind split elementari bind deep connect possibl split complex bind linear sequenc elementari navig result previou lower left part figur illustr rewrit bind oper algebra express figur rewrit simplifi queri composit push evalu sourc horizont navig type filter absenc type inform bad bind oper featur complex filter structur inform note introduct variabl remov project map att auction auction person bind tupl person set class person bind tupl person set class bind work work titl artist style work project size bind work work titl artist work bind work titl doc bind bind work titl doc split bind bind simplif bind artifact set class tupl class tupl auction person owner list artifact year creator price titl person bind bind project artifact set class tupl tupl auction person owner list artifact year creator price titl set join bind bind project artifact set class tupl tupl auction person class owner list artifact year creator price titl bind join figur algebra evalu strategi navig data graph pure semistructur system ad special achiev reason perform interest type inform data filter queri mix structur semistructur data semistructur queri structur data semistructur queri access structur tag variabl flexibl type filter illustr lower part figur retriev attribut person object type inform figur simplifi shown figur note resembl rewrit techniqu gener path express rewrit obviou bind oper push structur queri semistructur data partial structur xml assum user interest titl artist element artifact illustr lower middl part figur correspond project rewrit bind oper simplifi queri care chang type filter semant suffici condit hold type work instanc type filter xml view rewrit tree oper captur restructur semant queri view featur group sort typic oper tree rewritten sequenc sort nest map exist optim techniqu evalu tree remain costli appli larg amount data case user occur construct view import elimin intermedi ree oper result composit queri view definit time back evalu queri page left part figur present algebra translat view definit complex express correspond naiv evalu strategi view queri evalu result xml algebra equip rewrit express part figur due space sketch optim process essenti illustr arrow figur rid sequenc frontier view definit queri figur introduc instanti relationship filter lower bind tree rewrit sequenc simpl project renam deal oper standard rewrit project tree titl year price owner style lower part view definit tree bind queri select work titl bind tree select work work work titl work doc artist size doc histori bind bind work titl doc view definit figur figur optim appli artifact xml push project elimin branch simplifi bind xml sourc merg remain filter obtain final express note optim queri xml sourc full text subject section rewrit exploit sourc capabl queri process import techniqu distribut context push queri evalu extern sourc reduc process time sourc specif index similar fast access minim commun cost sourc cost middlewar limit system resourc requir benefit parallel introduc remot queri execut show descript sourc capabl section optim impressionist sold make match work doc algebra translat queri shown side figur transform sequenc project oper optim shown evalu sourc left string bind appli extract artist style element select pair titl sourc call retriev artifact inform aspect due oper correspond nest loop evalu valu variabl pass lefthand side side classic techniqu distribut queri optim obtain optim perform round rewrit round similar gave queri project simplifi bind sourc select push goal round rewrit push evalu sourc work requir explain section bind select trivial transform oql queri optim match bind oper wai capabl declar possibl push simpl bind xml document introduc select split bind match wai capabl descript step requir declar section connect select equal select step simpli figur round optim determin inform pass sourc base standard rewrit join summari present framework effici queri evalu xml integr system reli tree titl year price owner style lower part view definit work tree bind queri select titl price style work titl price answer view definit project tree titl price answer push oper bind work select work select bind titl wai bind select artifact set class tupl artifact creator price titl artist style artist doc size artist artist artist doc figur algebra translat optim gener xml algebra captur express power semistructur xml queri languag wrap structur languag oql sql algebra optim queri exploit type inform push queri evalu extern sourc work place context ya system develop bell lab inria xml version algebra evalu run stabl implement optim base heurist simpl linear search strategi consist rewrit round present section refer wiener lorel queri languag semistructur data intern journal digit apr delobel bulk load techniqu object applic relat data intern confer larg aug chu inform mediat mix acm sigmod confer manag page june demonstr beeri milo schema integr translat structur data intern confer theori lectur note comput jan beeri algebra semistructur data xml intern workshop web june extens markup languag feb wong data transform system biolog data sourc intern confer larg page sept thoma wimmer heterogen multimedia inform garlic approach research issu data page lo mar cattel object odmg morgan chaudhuri shim includ queri optim intern confer larg page santiago de sept scholl structur document queri facil acm sigmod confer manag page moerkott evalu queri gener path express acm sigmod confer manag page june mediat data acm sigmod confer manag page june cluet moerkott nest queri object base intern workshop program page york aug cluet moerkott queri process semistructur context suciu queri languag xml submiss world wide web aug suciu wareh increment evalu web site manag base de oct wadler data model algebra xml queri commun jan wadler xml queri experi exemplar commun sept fernandez suciu optim regular path express graph schema ieee intern confer data engin feb suciu queri optim presenc limit access pattern acm sigmod confer manag distribut system feder object relat object orient system solut advanc applic prentic graef queri evalu techniqu larg acm comput june yang optim queri data sourc intern confer larg page aug valduriez deal discrep wrapper function act de base de page sept lee adapt approach queri mediat heterogen inform sourc intern confer cooper inform system page june mchugh widom queri optim xml intern confer larg aug melton simon understand complet guid morgan resourc discoveri system comput network isdn aug ozsu valduriez principl distribut system prentic object fusion mediat system intern confer larg page sept object exchang heterogen inform sourc ieee intern confer data engin page mar ullman queri translat scheme rapid implement wrapper intern confer deduct objectori volum lectur note comput page dec pfeifer oct xml queri languag workshop xml queri dec simeon integr de sourc de comment phd universit de pari jan simeon cluet design issu xml unifi perspect draft oct xml schema part structur work sept valduriez scale heterogen design disco intern confer distribut comput page hong liu access heterogen data homogen integr mediat intern confer cooper inform system south june inform retriev applic servic definit protocol specif 
ML/1728 6 hybrid approach profil creation detect john daniel john inform technolog oper unit state militari academi abstract anomali detect involv behavior individu system recogn behavior norm paper describ preliminari result robust gener capabl machin learn method creat user profil base select subsequ classif command line argument base method belief legitim user classifi categori base percentag command period hybrid approach emploi begin applic expert rule reduc initi cluster data subsequ refin cluster locat competit network call learn vector quantiz learn vector quantiz nearest neighbor classifi record present network li distanc masquerad system requir record includ train set introduct detect defin problem identifi individu comput system author legitim access system privileg detect approach divid anomali detect system detect combin detect involv identifi pattern anomali detect pattern activ abnorm concentr anomali detect describ hybrid approach establish assign legitim user subsequ identifi individu attempt masquerad legitim user hybrid approach begin applic cluster techniqu assign legitim user cluster base frequenc command actual network environ reduc data genet refin relationship legitim user cluster variat competit neural network learn vector quantiz remaind paper organ section review anomali detect includ intellig statist system section describ hybrid section theoret underpin system section present experi section summari area futur work background previou work assumpt underli anomali approach detect behavior establish user act abnorm manner action user masquerad classifi behavior determin abnorm comparison user profil repres user typic behavior user base individu behavior typic behavior individu function group anomali detect system maintain model typic system behavior model system perform metric maintain time system oper normal manner likelihood intrud present system anomali detect techniqu appli problem detect research field detect formal public semin report den model detect system frequent cite model includ anomali record basic compon gener detect system public detect research appli wide varieti method detect activ earliest method detect applic statist method identifi activ earli system emploi method number project continu emploi statist method anomali typic combin method recent anomali detect method emploi wide varieti classif scheme identifi activ scheme rule induct neural network fuzzi set theori classic machin learn algorithm artifici immun system signal process method tempor sequenc learn challeng develop anomali detect classifi address featur reduct data impact perform data reduc effect system anomali detect approach address problem conceptu drift domain problem conceptu drift manifest behavior time effect detect system adapt chang recogn action adapt hybrid approach anomali detect introduct hybrid approach data approach detect base simpl legitim establish profil number type command emploi system discern command emploi system flag action potenti masquerad data build profil research consist individu command enter unix prompt set user user design command sequenc enter command line argument report time dai time individu command order creat user calcul percentag command emploi user total number command tabl data user assum command collect group command user repres address methodolog select number command exclud subsequ section notat notat paper user denot letter vector repres percentag command denot vector bold type number command dimens complet set data matrix repres number user group predetermin number command assum user classifi legitim masquerad ω goal research deriv assign user legitim masquerad base sole number type command note class legitim consist profil legitim user denot class legitim individu user profil denot number legitim user profil methodolog emploi methodolog build legitim user profil step subsequ explain paragraph step reduc data expert rule step cluster data step reduc data genet algorithm step refin cluster locat learn vector quantiz tabl data tabl entri repres percentag legitim user emploi base fix sampl command env kill dimens reduct expert rule facilit applic cluster reduc data mention data research consist individu command enter unix prompt set user group number command expert system elimin command command inform user profil retain expert system number rule fuzzi set membership tool captur semant human expert heurist fuzzi logic base notion object belong set unlik tradit set boundari set fuzzi fuzzi set theori notion membership set real number deal uncertainti real benefit fuzzi logic approxim solut complex problem exploit toler fuzzi logic good push knowledg acquisit bottleneck build expert system build fuzzi rule knowledg engin human expert describ problem semant build expert rule reduc import retain command typic frequent malici masquerad frequent normal user command frequent malici masquerad divid inform remot file upload compil support inform collect command includ tail remot access command includ file upload download support command includ ftp compil support command includ expert rule ensur command remov remaind rule combin easili comput statist fuzzi heurist remov command list output applic fuzzi rule reduc set command cluster algorithm cluster reli fact element share set expert rule analyz inform content command respect user command remov provid inform user rule screen invalid command data includ command symbol valid case erron argument list command difficulti ensur command standard unix command written execut local system elimin rule involv make command base easili comput statist reduct data user group set command instanc user user data set result instanc user shown tabl standard deviat instanc comput simpl appli basi rule elimin command deem statist command typic hacker normal user protect sampl data expert rule reduc number command classif learn vector quantiz theoret underpin profil creation lvq learn vector quantiz neural nearest neighbor classifi base kohonen work quantiz defin map broad rang input valu smaller number output valu input valu thought boundari set class output valu predetermin number node refer vector refer vector defin qr ℜ set refer vector denot divid refer vector number legitim profil refer vector pattern case legitim user profil strategi lvq effect train refer vector defin optim boundari class correctli posit refer vector lvq accomplish manner present train pattern input vector adjust posit select refer vector accord set learn paper train algorithm lvq attempt adjust posit refer vector input pattern refer vector categori nearest neighbor carri method kohonen argu refer vector approach centroid result voronoi tessel voronoi tessel partit ℜ disjoint polytop train pattern polytop refer vector nearest neighbor goal lvq approxim boundari voronoi approxim surfac bayesian classifi refer rigor mathemat descript lvq process prove converg algorithm asymptot condit lvq reli nearest neighbor scheme detect system advantag assum refer vector train lvq minimum distanc user closest refer vector defin ω ω min ω ω repres euclidean distanc refer vector user greater distanc user exce bound accept profil consid potenti masquerad import note methodolog note requir gener masquerad data creat profil train system legitim user measur deviat subsequ user profil problem neg data obviat lvq algorithm version lvq report literatur lvq algorithm attempt approxim surfac learn rule lvq algorithm slightli initi origin lvq algorithm refer futur work assess impact lvq algorithm classif process emploi learn law reward correct punish incorrect classif denot closest refer vector iter learn law closest refer vector belong class input vector closest refer vector belong class refer monoton function experi result gener masquerad data order gener realist masquerad data advantag fact malici masquerad typic command normal user identifi command automat includ gener probabl distribut command base characterist normal distribut randomli select record legitim user perturb actual number command emploi user random factor standard deviat due stochast natur gener record close resembl legitim user paramet valu hybrid exist kei paramet set user kei success lvq initi posit refer vector applic involv lvq emploi cluster basi initi posit refer vector requir user number cluster paramet valu includ number command assign legitim user number train iter lvq distanc threshold record classifi masquerad common test paramet set test system paramet set masquerad set data test repres result test present common paramet command user record train sampl record legitim user test sampl record anomali record predict predict lvq α lvq train test result emploi genet algorithm multivari linear regress fit function reduc number command kei command supplement command command typic total command note genet algorithm delet typic command recommend rule thumb time number multipli number command refer vector lvq train process test user masquerad distanc threshold actual anomali legitim anomali legitim figur result test interest record classifi record classifi cluster refer vector vector legitim user classifi cluster cluster uniqu cluster repres behavior test user masquerad distanc threshold actual anomali legitim anomali legitim figur result test half record misclassifi legitim closest cluster refer vector curiou random natur gener masquerad impli relationship larg number masquerad singl cluster slight degrad classif accuraci attribut number cluster test test test user masquerad distanc threshold predict actual figur result test command improv classif accuraci system system perform due comp ar small train set record larg number cluster experi conduct command lesser number result improv significantli analysi result anomali legitim anomali legitim surfac result approach merit futur detect problem gener profil exceedingli difficult problem definit behavior state test system high gener masquerad record case indistinguish legitim behavior paramet set approach robust brittl number iter lvq train process minu effect exact determin distanc threshold paramet differ test train easi determin threshold base classif train record paper preliminari result achiev applic varieti techniqu develop hybrid anomali detect methodolog detect appli expert rule reduc data perform initi cluster data final refin cluster locat competit network call learn vector quantiz initi approach show merit futur recal test system high standard achiev classif case misclassif rate futur research refin methodolog order classif rate misclassif rate attempt adjust number command user train sampl legitim user test sampl number anomali record lvq α number lvq train iter addit evalu methodolog gener appli data set refer ieee pp detect macmillan technic secur threat monitor jp anderson fort report number april ieee transact softwar vol pp foundat audit trail analysi detect comput ieee transact softwar vol pp sri id statist anomali present ieee symposium research secur comput session present ieee symposium research secur locat anomali detect nonparametr pattern recognit present comput secur applic san unusu program behavior statist compon sri menlo tech report detect appli machin learn solari audit present comput secur applic implement scalabl detect system protect network present darpa inform surviv confer hilton head network detect project vol emerald present usenix workshop detect network santa realtim anomali detect induct gener sequenti present ieee symposium research secur lo softwar architectur languag audit trail present european symposium research comput vasant intellig agent detect present ieee inform technolog base detector recogn present ieee workshop inform assur west detect neural present advanc neural inform process system map present nation inform system secur detect soft comput present secur paradigm rhode data mine base approach detect distribut recent advanc detect intern octob vol vol ed pp framework construct featur model detect acm transact inform system vol unix present ieee symposium research secur commun vol pp anomali detect distribut intellig ieee vol pp sequenc learn data reduct anomali acm transact inform system vol pp detect system call ieee vol pp acquisit depart inform comput scienc data assembl conjunct rutger sj uncertainti ieee transact son vol pp fuzzi control prentic expert principl publish learn vector quantiz pattern recognit technic associ memori neural theoret foundat analysi ieee nonparametr classif learn vector quantiz ph pattern recognit neural benchmark ieee intern confer neural san pp classif learn vector technik mr cluster analysi applic academ 
ML/1869 6 attribut grammar genet neural network syntact constraint genet program roger queen ontario present workshop evolutionari comput attribut grammar grammar augment assign semant attribut symbol grammar product rule specifi replac evalu symbol attribut attribut grammar class neural network structur explicit function organ provid constraint genet optim preserv syntact correct genet tree semant give broad overview research attribut grammar basic current idea futur direct research attribut grammar neural network attribut grammar appli techniqu repres gener class neural network approach involv principl design context free syntax attribut manipul interpret nontermin termin symbol grammar repres neural network modul attribut neural network modul defin set neuron hidden neuron design connect product rule design respect intern integr modul repres symbol hand side product modul repres left hand symbol product rule creat connect design input output node hand modul input output node left hand modul hidden node hand modul process gener network attribut grammar involv step tree gener start symbol applic product rule term deriv tree attribut symbol deriv tree evalu attribut manipul rule grammar product gener fulli attribut tree specif neural term termin refer symbol grammar discuss term intern node leav refer placement element genet tree clarifi differ genet program terminolog network structur extract attribut root interpret produc function neural network neural network randomli initi train data term approach network gener attribut grammar encod figur present grammar gener layer network input output symbol refer neuron object type creat uniqu id hidden connect attribut product rule product rule specifi syntact rule set attribut manipul attribut left symbol set equal attribut symbol neuron input output hidden output hidden connect input output neuron neuron input input output output hidden output input hidden connect connect input neuron neuron input input output output hidden hidden connect connect node node neuron node input node output node hidden node connect node figur grammar layer network deriv includ recurr kohonen arbitrarili layer classic backpropag network research continu extend front interest consequ research neural network architectur model similar differ common framework grammat descript network architectur suggest techniqu integr multipl architectur form hybrid class attribut grammar genet program attribut grammar optim neural network genet program wai attribut grammar form genet encod class neural network attribut grammar properti exploit design applic oper genet program techniqu grammar specif neural network deriv attribut final function neural network specif neural network inform function organ structur consequ grammar design principl previou gener deriv tree correspond structur compon network deriv tree indirect encod neural network intellig genet manipul made design genet genet oper ensur offspr produc valid encod provid deriv requir genet oper form offspr deriv tree gener scratch applic attribut grammar product solut constrain genet oper oper entir ensur root symbol replac root nontermin symbol solut relat work crossov genet program hayn figur demonstr constrain crossov deriv tree form grammar figur box replac box form deriv tree syntact valid grammar genet import genet manipul semant genet tree unpredict result function genet search random walk properti deriv tree correspond structur modul ensur genet oper correspond semant substitut root symbol effect substitut modular structur effect remain neural network structur concern connect made input output node modul figur constrain crossov genet oper design access rule grammar mutat oper identifi symbol present select point tree gener appli grammar rule symbol initi symbol attribut grammar addit benefit search good neural network solut attribut grammar produc space architectur explor requir genet program design random applic product rule creat popul result random fine structur neural network solut attribut grammar abil class neural network basic structur random choic product reflect larg architectur characterist reflect smaller structur detail approach evolutionari optim neural network integr varieti principl singl includ genet modular neural network grammar base genet encod neural network boer jacob syntact constraint genet program hayn node seq par node node map map seq seq par node node map node node seq seq node node node current work neural network attribut grammar made slightli interest assign probabl product rule random gener rule applic select probabilist effect gener process encourag structur form final neural network note class network repres grammar frequenc network gener genet effect rel frequenc gene popul probabl combin mutat natur constrain type mutat occur frequenc extrem structur popul remain low evolut extrem case exhibit high fit applic genet oper mutat crossov requir select intern node deriv tree neural network make current popul intern node repres close start symbol larg structur modif select node correspond close termin grammar result small structur chang current genet oper select provid abil bia scale structur investig dynam set probabl shift scale structur genet algorithm investig reinforc learn strategi adjust reinforc signal deriv fit measur newli gener network conjectur effect dynam manipul probabl consequ genet program seek structur earli gener finer structur gener result compar similar find poli standard genet program futur direct expect attribut grammar system lead cultur evolut spector evolutionari comput complex learn compon neural largest proport comput time dedic evalu fit individu evalu fit neural instanti genet code undergo learn environ end neural network great deal acquir inform typic inform transmit train start scratch conjectur cultur evolut process inform learn prior individu transmit quickli evalu fit individu learn problem start point transmiss cultur inform involv direct transfer neural network structur indirect transfer behavior inform neural direct transmiss difficult perform gener difficult identifi compon individu share ident structur difficult support claim compon perform similar function suspect intermedi attribut quickli identifi match structur compon individu parent permit direct transfer structur inform ident compon form similar locat genet tree perform similar function final network refer social structur commun agent meme fourth annual confer evolutionari program mit gv evalu left commun acm ig modular artifici neural leiden tech report origin modern mind harvard press definit modular neural adapt behavior syntact constraint genet chapter ke jr pj angelin advanc genet program mit press evolut modular neural network neural da rl inherit strongli type genet chapter ke jr pj angelin advanc genet program mit press ts ra gener attribut grammar ieee intern joint confer neural ts properti attribut grammar annual genet program phd student workshop juli wisconsin neural net architectur hierarch genet neural network genet algorithm graph gener complex semant mathemat system theori jr genet program automat discoveri reusabl program mit press dj type genet bbn tech report wb analysi schema propag disrupt genet seventh intern confer genet transmiss inform genet annual genet program confer 
AI/2906 2 random probabl function jon williamson march abstract random probabl function requir mont carlo simul expert system test give empir method gener function practic solut interest philosoph question design expert system import problem eld arti cial intellig give gener methodolog diagnosi construct comput expert rang area diagnosi hepat nding circuit board arti cial expert make critic test reliabl ascertain approach test stand empir experi mont carlo simul empir test empir data readili ascertain reliabl diagnost expert system requir amount symptom correct larg number past case problem compound test gener diagnost methodolog case requir rang sucient data test isnt real data good idea reliabl expert system natur strategi gener data random test simul data markov chain mont carlo provid algorithm sampl data random method sampl data probabl appli rst determin random probabl function simul strategi gener random probabl function sampl produc random test expert system methodolog sampl theori branch distribut probabl function literatur question subject probabl function adopt face ignor object function consensu function techniqu rang gibb sampl recent exact method coupl past extens literatur instanc al entropi subject issu applic problem address subject approach concern singl function convei inform simul requir gener function repres object includ highli inform function progress made issu random probabl function work al approach motiv larg conceptu contrast develop hope altern perspect advanc debat conceptu problem test expert system methodolog al search distribut probabl function bear goal gener random probabl found randomli sampl natur distribut approach base kei natur distribut natur joint probabl distribut pm variabl give joint probabl distribut pm distribut pm ident pn natur distribut variabl al give intuit expect natur distribut probabl function render variabl depend fairli natur distribut expert system reason valu variabl valu variabl diagnosi expect variabl convei inform binari valu variabl expect symmetr invari transposit peak probabl encount everydai life close linguist categori divid popul half descript observ uniform distribut probabl function fail render variabl develop model satisfi constraint roughli natur probabl distribut distribut jn probabl larg logic sentenc proposit variabl independ ident distribut al full descript model empir approach situat restrict context expert deal arbitrari causal relat variabl focu outcom toss singl form gener random probabl function restrict toss independ ident problem reduc random select probabl toss yield head empir approach examin sampl estim gener random valu base observ distribut expect experi probabl similar data consist outcom roll die observ roll case expect empir approach lead natur distribut sharpli peak clear approach yield gener natur distribut wrong gener random valu distribut peak gener resembl gener experi abil context motiv empir approach gener problem simul expert system test examin real context expert system methodolog vision analysi observ paramet joint distribut determin form natur distribut paramet main hurdl approach pragmat gener random probabl function sampl requir paramet probabl function binari form distribut prohibit sampl costli term comput worst case jump hurdl observ gener probabl function random expert system test suce gener repres spread probabl function gener similar observ test problem solv solut base theori constrain bayesian network exibl formal control probabl function constraint reduc spread probabl function gener reduc number paramet requir probabl function comput time requir sampl function trade repres spread speci cation comput rest section give introduct constrain bayesian network show network appli problem gener random probabl function constrain bayesian network domain cn nite valu liter express form state conjunct liter bayesian network consist direct acycl dag node cn set valu jd liter involv node state parent ng independ parent node probabilist independ set node descend jd jd bayesian network suce determin probabl distribut node cn probabl function cn repres bayesian network complex number speci ed valu depend structur graph worst case complet dag arrow complex exponenti graph number dramat reduc variabl discret graph requir speci tree requir complex graph parent node requir complex complet dag requir complex time calcul condit probabl state depend structur graph polynomi time propag worst case time exponenti gibb sampl ectiv sampl bayesian time gener sampl depend crucial time calcul probabl form structur graph constrain bayesian network exploit featur place constraint bayesian network limit complex time produc sampl instanc constraint node parent complex linearli constraint path node requir valu calcul ecient determin probabl distribut node probabl calcul determin proport case approxim distribut determin bayesian network subject constraint network ord closest approxim measur log rang atom state state involv variabl jd instanc redund valu determin addit network satisfi arrow weight order weigh arrow node parent weigh arrow condit mutual inform log jd jd jd wai constrain network maximum close maximum simplest greedi start graph arrow stage weigh arrow addit ensur dag structur constraint remain sati add maximum weight arrow exist spawn graph ad arrow previou constantli prune number graph elimin longer maximum weight stop algorithm arrow ad gener random probabl function plan sampl real approxim distribut constrain bayesian network obtain natur distribut paramet approxim network gener probabl function random construct constrain bayesian network distribut paramet constraint limit bayesian network parent node network compact easi sampl gener network random determin number node number valu node determin graph node sati es structur constraint final determin speci cation gener jd liter parent state procedur data measur sampl constrain network approxim distribut number number valu graph speci cation concret sampl select machin learn repositori distribut number node network figur show distribut sampl thing import note ensur sampl fair indic size expert system appli kind distribut smooth requir valu data render advantag random simul test system assess cope extrem justi cation perform algorithm merz figur distribut number node con gurat sampl data smooth import requir distribut number valu node assum number valu node independ number valu care node sampl variabl domain partit nite number choic indic choic futur applic practic sampl approach figur distribut machin learn binari order obtain distribut graphic structur measur number parent node figur sampl distribut gener random graph subject distribut node determin number parent distribut reorder parent predecessor make choic order break constraint node parent order figur graph resampl number parent dag consist sampl node award parent final obtain distribut probabl speci er jd distribut di er number valu node number order call order figur distribut number valu figur distribut number parent hj figur graph figur speci parent parent figur distribut node valu parent figur figur parent figur show number valu skew distribut peak empir approach involv step repres sampl constrain network approxim measur select characterist network gener constrain network random base distribut characterist sampl constrain network gibb sampl approach bene ts abil narrow class fact suitabl choic constraint lead mechan practic implement issu section make remark put forward idea research figur speci parent figur speci parent figur speci parent conceptu intuit empir rule bear conceptu intuit al variabl interdepend empir account fact constrain network link node arrow probabilist depend exist node mutual inform kind measur condit probabilist greater node link framework gear repres depend relationship node figur expect peak present distribut roughli symmetr symmetri achiev choic probabl interv close left open nal interv close empir account symmetri due invari renam fact probabl sampl node parent expect condit probabl jc peak strongli node reveal signi inform vice versa figur con rm intuit figur show ad parent reduc uncertainti higher peak mention node valu trend remain skew empir procedur satisfi condit experi gener random constrain network binari gener subnetwork node distribut binari node figur empir approach case random network node repres figur call univari natur condit requir distribut singl node subnetwork univari distribut fairli stabl figur show vari case gener bayesian network uniformli select graph probabl uniform figur margin distribut peak rst indic condit sati ed evid requir con dent order pursu conceptu approach import strong conceptu constraint elimin candid natur distribut dont failur problem empir approach real natur distribut satisfi goal empir approach close real intuit principl hold test approach al set logic framework variabl proposit binari valu constrain network variabl nite number valu question approach extend variabl continu variabl figur uniform distribut question give test success extens probabl densiti function singl variabl valu real number real number written case suce determin function give function thought densiti function distribut signi digit valu variabl type data scale invari depend choic unit measur mean signi digit data base invari depend base data scale base invari rigor formul sati es benford law origin due signi digit number distribut ith signi digit ith digit log expect lead digit number percent time altern deriv benford law gener probabl function natur distribut distribut expect function turn signi digit function show benford law hold mix sampl randomli gener function scale invari base invari substanti empir evid claim benford law hold aggreg sampl hold sampl individu start ten digit occur equal frequenc evid make logarithm notic faster rst page wear logarithm tabl type data notic aggreg ect aggreg data gener random expert system test satisfi benford law test test aggreg data directli mechan gener probabl function random yield scale invari base invari subject probabl deal object probabl interpret frequenc subject interpret interest expert system perform random gener agent belief function knowledg natur distribut probabl function repercuss subject probabl intuit plausibl absenc relev knowledg domain agent set belief function domain expect probabl function natur distribut repres belief state face ignor agent variabl valu intuit belief function expect random probabl function set node question mind extent justifi plausibl give dutch book argument identi cation belief agent bet di erent belief long run tend lose monei oppon set stake extent identi cation di er principl set belief face jayn maximum entropi principl set belief function probabl consist refer empir evid found recent exposit found maximum entropi log rang atom state expect di er maximum entropi solut face total situat object natur distribut approach random probabl conceptu empir approach empir approach er practic mechan gener random probabl function easi sampl exibl appli di erent gener random probabl function domain sampl constrain network domain fact probabl function sampl xed natur distribut concret mean notion random render approach philosoph appeal interest philosoph question relationship conceptu empir approach approach statu conceptu principl knowledg random uenc ration belief refer merz cl blake cj repositori machin learn depart inform comput scienc al wr richardson dj chain mont carlo method chapman hall impli benford american mathemat societi page statist deriv signi statist scienc page logic simon frequenc di erent digit natur american journal mathemat page art human research board fund research al jb pn gm structur probabl function natur intern journal fuzzi reason stochast arti cial intellig page reason intellig network plausibl san morgan kaufmann jon discret probabl distribut bayesian report pai jw 
ML/443 6 workshop learn text june build intellig agent approach shavlik tina dayton street wisconsin abstract present evalu infrastructur rapidli easili build intellig softwar agent task design center basic highli accur standard heurist search lead effici retriev inform approach user tailor system behavior provid approxim advic function advic map neural network implement function subsequ reinforc web dead rate retriev page user provid refin function agent architectur appeal middl ground program languag system sole learn user prefer user rate page present case studi provid simpl advic special system empir studi demonstr approach lead effect homepag finder lead commerci web search engin introduct describ evalu implement system substrat creat intellig agent web approach base idea commun kibler ourston moonei towel shavlik user person interest prefer languag design discuss aspect content structur web page instruct neural subsequ refin train exampl wisconsin adapt web assist idea reinforc learn automat creat train train exampl design import advantag produc agent describ show easili instruct creat finder empir studi finder present result support claim version outperform finder search engin hotbot experi demonstr improv perform train exampl automat gener system descript heart neural implement function guid system wander web judg page encount user program function provid call guid wander score page shavlik call program languag advic underli system blindli follow refin advic base system experi network larg input vector set express advic user person advic essenti attent small subset make learn feasibl cancer research stock analyst express interest advic regularli monitor relev web site articl interest type potenti user develop build intellig agent top peopl result agent phrase user case type user provid advic underli neural imagin type user indirectli special interfac type user creat scenario section tabl descript initi neural network creat read disk resumpt previou view process convert advic neural network analog compil tradit program machin system compil instruct intermedi languag express neural network import advantag automat refin base feedback provid user web tabl algorithm save disk previou creat neural network read user initi advic start ad url search initi search queue url queri user chosen set web search engin site execut concurr process independ process search queue empti maximum number url fetch evalu network score high insert sort list page found score improv predict network evalu hyperlink network score link insert url search queue fit bound independ process user addit insert neural network independ process person rate fetch rate creat train neural network basic oper heurist function provid sort heurist sole find goal collect page rate highest user seed queue page fetch set start url provid simpl queri convert url subset select search engin site mention tabl user depth limit put upper bound distanc system wander initi url fetch page initi predict fetch base content page link fetch analyz system estim page user differ estim constitut tempor differ error backpropag william improv neural network addit method automat creat train user improv neural network wai user provid addit advic observ system behavior invok thought good addit instruct accept advic augment neural network time simpli add network addit hidden unit repres compil techniqu effect demonstr shavlik task provid addit hint rapidli drastic improv perform provid advic relev paper experi evalu increment provis shavlik testb show algorithm robust quickli learn ignor user rate page mechan provid train exampl bp user unabl articul system page provid score standard methodolog investig billsu discuss aspect articl improv neural result user provid advic person easi simpl advic requir larg number label exampl order learn pure induct empir support claim topic experi progress neural network mean mechan process arbitrarili long web page input vector borrow idea nettalk rosenberg basic unit word letter nettalk slide window featur repres page defin respect current center window defin score page highest score network produc page hyperlink comput slide network hypertext hyperlink case window start center word mean nearbi word hypertext fill window posit overview advic languag turn repres web page construct advic languag input featur extract html plain constitut primit advic languag descript basic briefli discuss complic languag construct creat basic extract featur web page standard text inform retriev model left side fig illustr word order lost vector record word present scale number occurr properti tfidf inform retriev system discard common word word root form greatli reduc problem perform preprocess step sole richer preserv inform advantag structur html document fetch page format partial shown line tabl augment local illustr side fig bag word word bag page left current hyperlink window insid current section titl parser web page record titl parent word standard construct indic tabl caption head bag word grandpar current window nest word clarif order web page url page content refer origin web page sampl page space rent stop word remov stem standard approach aspect local space rent page sampl www page sampl word titl word window space rent word url page www sampl page space rent slide window figur intern repres web page tabl sampl extract input featur url case attempt reduc addit word repres fix posit obviou case posit slide repres word fix section due import role special repres field delimit server portion url www aaai org featur repres web rang current version tfidf due manner compil advic neural design lead larger number input assum typic vocabulari ten thousand order repres input vector record featur advantag import aspect neural network repres absent word input featur plai role phase neural weight sum bp due partial deriv involv input featur relat word posit input vector includ length date page creat page server provid window insid html size word word mention advic present learn system hope larg space input featur deal input featur sole learn label exampl advic mean user indirectli select subset featur space huge implicit input vector indirectli select featur involv word advic full input space weight input featur advic high weight valu potenti word mention advic impact network bp complex predic tabl complic predic defin term basic input featur advic finder experi tabl construct tabl satisfi list word basic instruct output tabl sampl advic joe smith home strongli suggest show page joe smith joe smith index home homepag html strongli suggest link strongli suggest avoid show page neural network entri tabl illustr advic map neural network assum seek joseph smith home page intent rule system slide window page plausibl variant person phrase condit larg weight sum output unit network accomplish variant algorithm shavlik rule tabl compil posit treat separ slide constraint itl predic true activ connect referenc input unit newli creat hidden weight set bia hidden sigmoid activ requir predic true order weight sum input exce bia produc activ hidden unit addit link ad hidden subsequ standard link hidden unit output unit weight determin strength rule consequ interpret phrase show page variant rule case studi possibl smith middl initi home possibl titl form page rule show piec advic homepag find compil input true word nth end current hyperlink weight sum substanti treat url separ rule show advic follow link show negat avoid instruct neg weight neural network experi section present case studi illustr effect creat special agent top system task chose task creat finder etzioni system techniqu call dynam refer filter output web indic gener guess url candid found wrote simpl interfac layer top relev inform person home page middl miscellan partial url wrote short program read field creat advic wrote gener advic rule relat slight variant middl special task creat initi gener advic spent part dai advic describ technic comment fulli understand detail experi user retract advic neural network advic ad request find home page learn gener homepag find accomplish crude variabl bind mechan accept instruct word bound input vector field assign queri person ord altern ord ord write advic find featur rule tabl written marker specif motiv methodolog randomli select peopl aha list machin learn reason research run experi evalu reduc comput load limit peopl unit state peopl randomli pick train remain test set manual ran system manual advic advic evalu train train set judg perform task find provid advic discuss import note provid advic specif ai option effect amount execut time accuraci result chose small number maximum number page fetch includ queri initi search maximum distanc travel page return search engin start provid person aha web partial standard exampl common variant convert initi queri search engin mention earlier compar perform perform search engin perform experi etzioni provid web interfac ran hotbot condit set perform special hotbot search aha page queri variant provid hotbot disjunct person requir variant person word queri initi send search engin page hotbot assumpt peopl result return search engin peopl link home compar url provid aha exact comparison content fetch page content page link aha page run fetch url server match aha prevent aha page bp learn perform experi refin automat creat train exampl discuss ran control result discuss tabl list result report percentag report averag ordin posit page hotbot return sort list result provid strong evid version special finder ad simpl produc finder proprietari creat accuraci differ row suggest effect result suggest perform differ confid level cost approach fetch analyz web longer speed ignor question fetch charact web capsul summari search engin relev statist find home averag ninth page fetch tabl empir result system found rank bp bp person search hotbot gener relat work syskil webert billsu webwatch mitchel web agent machin learn techniqu bayesian classifi hybrid learn unlik system unabl accept simpl provid lead learn rate manual visit web page current futur work plan valid claim appeal middlewar creat addit person electron newspap email filter plan continu expand advic languag build abil inform synonym wordnet knowledg text add capabl automat creat plausibl train exampl observ action made user ordinari present evalu appeal approach creat intellig agent web central aspect design system core mean agent built top adapt mean user type defin tailor result agent match person prefer rate inform retriev argu creat softwar agent involv abil direct program provid set approxim abil accept automat train exampl due larg unstructur natur size hybrid approach appeal sole base program languag reli user rate larg number web page case studi present support claim refer tour guid world wide web proc creat reinforc learner machin learn lexic english commun acm theori combin analyt empir method artif intel util knowledg induct learn machin learn identifi interest web site proc learn error natur develop automat text retriev scienc parallel network learn pronounc english text complex system dynam refer case studi homepag domain proc sixth intern world wide web intellig agent approach workshop learn text categor learn predict method tempor differ machin learn artifici neural network artif intel 
ML/920 6 submit acm autonom agent synthet agent system bayesian model human interact barbara oliv alex pentland vision model media laboratori ma usa abstract build statist machin learn model real data frequent encount difficulti limit amount train data compar specif learn architectur order deal problem develop synthet agent train system develop flexibl prior model recogn human interact pedestrian visual surveil task demonstr abil prior model accur classifi real human behavior interact addit tune train introduct solut develop applic agent system steadili number agent appli domain comput game interact inform retriev user interfac electron industri process control paper agent prior model sourc train data system model human specif synthet agent mimic human interact model recogn goal system accur interpret behavior interact pedestrian scene ing approach model interact statist learn techniqu teach system recogn normal behavior common interact graphic model hidden markov model coupl hidden markov model model classifi human behavior offer dynam time train clear bayesian semant individu interact coupl gener process major problem statist model rare limit number exampl behavior train model compar specif learn architectur critic problem gener model captur prior knowledg human behavior order deal problem creat synthet agent train system develop flexibl prior model recogn human interact pedestrian visual surveil task select prior controversi open issu bayesian demonstr abil prior model accur classifi real human addit tune train major emphasi effici bayesian integr prior knowledg synthet prior evid data paramet approach straightforward flexibl techniqu design requir strong analyt assumpt made form prior experi found combin synthet prior limit real data easili achiev high accuraci recognit interact system robust case exampl behavior exampl visual surveil system overview visual surveil system emploi static camera wide watch dynam outdoor scene extens activ camera straightforward plan comput note prior form markov model vision system segment move object learn scene scene descript method variat learn accur discount move object descript track temporari multiobject meet kalman filter track object color veloc tempor order stream data obtain behavior descript detect interact object figur depict process loop main function unit ultim system extend descript comput vision system direct reader comput vision input modul detect track move object move object output featur vector describ motion spatial relationship nearbi move object featur vector constitut input stochast behavior model hmm vari structur depend complex classifi perceiv behavior visual evid pedestrian detect track focu attent model expect expect imag track interact detect recognit detect foveat input model percept select attent predict predict attent model expect vector featur visual input learn figur process loop synthet behavior agent develop framework creat synthet agent mimic human behavior virtual environ agent assign behavior interact gener interact behavior kind individu behavior paramet virtual environ model basi real pedestrian scene obtain measur typic pedestrian movement agent architectur dynam system consist number agent perform specif behavior set behavior system start time move discret forward time agent disappear scene agent follow path direct walk random speed random instanc time slow speed stop chang direct independ agent scene precondit satisfi specif interact agent place agent perfect knowledg includ posit agent loss system gener prior model synthet data agent interact agent make depend type locat locat agent scene script behavior priori knowledg kind place agent behavior determin perceiv contextu rel posit path direct repertoir behavior trigger event agent decid follow proce path speed progress reach walk path agent adapt mutual speed order continu advanc exit scene attent focu attent interest simul human interact visual surveil system continu observ monitor attent mechan record situat eventu lead interact specif event signal system interest happen chang direct agent place agent nearbi radiu attent simul start dump agent featur vector figur illustr mechan path window agent attent figur attent window synthet agent environ agent orient veloc data featur vector construct consist deriv rel distanc ff align magnitud veloc note featur vector invari absolut posit direct agent environ agent behavior agent behavior system structur hierarch primit simpl behavior agent perform walk chang chang start walk higher level complex interact behavior simul human interact specif tempor success simpl behavior event signal chang current simpl behavior agent perceptu system detect event react agent close agent fall attent radiu agent agent agent reach agent occupi roughli posit agent start agent talk amount chat timer expir agent start move signal agent timer agent timer stop dont stai stop forev agent timer start expir moment agent start move experi report consid interact behavior illustr figur reach walk agent happen path walk direct agent decid reach speed order reach agent slow walk speed meet separ agent path opposit direct close realiz slow final stop chat talk independ meet agent behav talk decid continu agent direct follow chang direct order meet continu agent start path close decid interact agent wait reach direct order wait agent chat time decid chang direct order meet separ interact decid separ independ note figur portion agent trajectori featur vector fell attent window simul depict assum interact occur radiu agent record trajectori proper design interact behavior requir agent knowledg posit synchron success individu behavior activ agent figur illustr timelin synchron simpl behavior event constitut interact interact happen moment time provid interact satisfi speed walk durat start end action vari highli high varianc quantit aspect interact confer robust learn model tend captur invari part interact invari reflect natur interact environ behavior hmm section describ framework build appli model individu behavior interact order build effect comput model human behavior address question knowledg map comput dynam deliv consist interpret statist direct acycl graph probabilist infer network provid effici solut rel distanc deriv rel distanc veloc magnitud align agent trajectori featur vector rel distanc deriv rel distanc veloc magnitud align agent trajectori featur vector rel distanc deriv rel distanc veloc magnitud align agent trajectori featur vector figur trajectori featur vector rel distanc deriv rel distanc veloc magnitud align agent trajectori featur vector rel distanc deriv rel distanc veloc magnitud align agent trajectori featur vector rel distanc deriv rel distanc veloc magnitud align agent trajectori interact featur vector interact figur trajectori featur vector chang chang direct interact behavior chang speed speed time walk walk reach agent direct path agent close speed walk walk event follow start walk start walk walk walk stop chat chat agent close leav agent expir timer path opposit direct reach agent chat approach separ approach speed walk walk time event direct start walk chang direct walk chat chat stop walk expir timer opposit direct path speed reach agent chat agent close approach approach speed walk walk time event agent close approach walk walk chat chat path leav agent expir timer agent approach path reach agent chat chang separ walk stop wait walk time event walk walk start walk chat approach chang direct reach agent chat timer expir path speed agent close path agent approach chat time walk stop wait stop chang direct event walk figur timelin complex behavior term event simpl behavior figur typic imag pedestrian plaza question hmm view simpl case tempor pin dag paper human behavior examin gener pedestrian walk open outdoor environ goal develop composit analysi observ behavior term state transit state time manner state correspond common notion human immedi applic wide rang site view situat figur show typic imag pedestrian scenario hidden markov model popular probabilist framework model process structur time clear bayesian effici algorithm state paramet automat perform dynam time warp hmm essenti quantiz system configur space small number discret probabl transit state singl finit discret variabl index current state system inform histori process futur infer reflect current state variabl graphic hmm depict time figur interest system multipl interact merit composit variabl typic case system structur time space singl state markov model problem order model interact complex architectur direct reader extend descript observ state state observ figur graphic hmm time technic perform compar analysi hmm model human interact extens basic markov model gener memori system provid composit state time interest system composit state simultan state variabl exact solut extens basic hmm chain intract case approxim techniqu exist exact solut case interact case coupl hidden markov model model interact case correspond individu human architectur state chain coupl matric condit probabl model causal influenc hidden state variabl graphic shown figur graph state time depend state time chain influenc chain causal link result section describ experi perform analyz real pedestrian data synthet model train data site data collect preprocess person detect track system obtain blob featur person hour video exampl type meet behavior detect process featur vector comput vision process modul centroid person kalman filter state instant consist repres filter compon gaussian fit blob color space vision system hz sgi comput filter data cutoff filter comput pair nearbi person featur vector consist deriv rel distanc jv norm veloc vector ff align trajectori person typic trajectori featur vector meet continu separ behavior shown figur type behavior displai figur synthet agent note similar featur vector case figur trajectori featur vector interact meet continu separ behavior human behavior model evalu result model meet continu meet split follow interact versu interact detect test perform hmm perform omit report result order evalu synthet agent environ train type behavior model learn synthet agent environ directli appli real data addit train tune paramet posterior behavior model train start point synthet model exampl interact data specif site recognit accuraci summar tabl noteworthi train recognit accuraci real data result demonstr abil accomplish extrem rapid refin behavior model initi prior model final roc curv posterior displai figur interest result experi high accuraci obtain test priori model obtain synthet agent simul fact priori model transfer real data demonstr robust approach show synthet agent train develop model type behavior avoid problem limit amount train data appli model real human behavior addit paramet tune train agent paramet sensit order evalu sensit classif accuraci variat model train set model chang paramet agent dynam factor perform alter model turn virtual case sensit peopl rel rate movement test real pedestrian data prior posterior accuraci real data roc curv real pedestrian data figur left accuraci priori model test real pedestrian data entri row interact remain entri classif accuraci interact behavior interact reach walk meet meet continu roc curv real pedestrian data summari paper evalu agent prior model sourc train data system model human specif synthet agent mimic human interact model recogn main element bayesian perceptu synthet agent comput vision system mathemat model framework recogn human behavior interact visual surveil task system combin inform close feedback compon emploi statist bayesian approach statist learn hmm model behavior interact found superior hmm term train effici classif accuraci synthet agent train system creat order develop flexibl interpret prior behavior demonstr abil priori model accur classifi real behavior addit tune train fact special limit amount train data acknowledg sincer michael toni jebara matthew brand insight comment refer wl guid literatur learn probabilist network ieee transact knowledg data lawrenc tutori hidden markov model select applic speech vol pp matthew alex hidden markov model complex action ieee matthew hidden markov model model interact submit neural novemb model recogn human novemb rk percept passiv pp bayesian comput vision system model human gran januari david tutori learn bayesian tech rep microsoft june lawrenc saul michael chain hidden markov gari david tk vol 
IR/738 5 document classi cation base hyperlink document semant kuo wong depart comput scienc engin hong kong hong kong abstract basic web document set hyperlink point relat document hyperlink document provid inform relat web document analyz hyperlink document identi ed algorithm classifi web document subset base hyperlink document content repres document identi ed subset base similar de nition repres search relat document achiev introduct www grow decentr document web lack logic organ enorm number web document make manipul oper web document di cult size web document set analyz web document web document divid disjoint set base document content problem web document classi cation web document categor web document set hyperlink point web document set hyperlink provid inform web document algorithm partit set web document base network hyperlink structur present similar de nition document base document content measur similar document partit de nition similar prune document order maintain consist document subset paper organ algorithm preliminari classi cation hyperlink introduc section similar de nition document classi cation process present section author support part hong kong grant result section section document partit hyperlink algorithm web document organ pebl vsm categor algorithm base classi cation paradigm author classi er document classi cation concept mobil agent appli classifi web document page interest reader refer refer detail algorithm inadequ requir larg number featur learn improp distanc measur due variat document length word discuss web document categor algorithm base hyperlink analysi similar de nition comput discuss web document set hyperlink set web document node hyperlink direct set web page transform direct graph page correspond node hyperlink transform direct edg de nition de nition direct link de direct link node node base de nition de ne import classifi document di erent class de import document de ned de refer web document de ned import de ne import document set document document mean document import document high document refer document refer high refer import document high page low base de nition document classi ed di erent type document discuss topic partit subset de ne set node hypergraph di erent type document subset document classi ed relat document relat node set relat di erent type node identi center node termin node center node point termin node node direct edg node edg point node relat node hypergraph link subset speci topic type node unrel node de nition node base de nition de node refer node consid center node de node refer partit subgraph belong node consid unrel node de node consid node set center node set relat document gener manner set node center node edg point put set node ad subset termin node unrel set node node point ad subset algorithm termin unrel node node encount edg center node relat node di erent unrel subset combin base set center node subset potenti repres identi ed set potenti repres minim set center node point largest number document subset greedi algorithm center node largest number document subset select node center node point point set node potenti node set potenti similar comput focu document subset document retriev import center relat unrel fig gener hypergraph web document subset due unstructur properti emploi object exchang model data model manipul web document oem label tree structur refer detail oem web document construct web document follow basic de nition hypertext markup languag web document transform oem data record html document transform oem record gure hierarchi tag html document set tag html document identi cation document structur hierarchi ofth document content set repres import titl topic head structur tag tag set structur tag assign id level hierarchi html document assign id assign id larger higher level hierarchi base set structur document set node organ hierarch transform algorithm convert web document preliminari oem record shown gure tag edg label section head intern node content convert intern node content edg label remov origin edg label origin oem record gure gener similar de ned base oem record transform de ne document node edg label similar structur similar similar document linear combin similar node similar node oem record transform web page set word origin content document de ne individu simpl html page write html write html difficult html tag write html difficult simpl html page html tag write html fig transform html oem format oem format node similar node node similar individu node similar match word content node de ning individu node node similar oem record de ned base node similar de ned leaf node edg label similar comput edg label similar base edg label branch oem record branch de ned edg root leaf de nition edg label similar individu edg label branch label match label branch charact arrow edg label edg label similar note sequenc edg label comput edg label similar edg label similar edg de easi de ne edg label similar structur similar de ne similar base structur oem record simplest tree leav node intern structur similar tree structur similar cnt di erenc cnt cnt di erenc cnt cnt di erenc cnt di erenc cnt parent parent structur similar de ned node similar repres select base similar de nition de ned previou similar oem record de nition similar similar potenti repres member comput potenti repres maximum averag similar document subset select repres repres averag similar document higher inform select simpli repres result report result show ectiv algorithm compar algorithm classi cation algorithm set web document set data set web document gather base di erent topic set set document total unrel set document extract set word highest occurr document extract featur vector classi cation result compar base recal tabl result algorithm quiet set noisi set environ classi cation number comput recal classi cation set paramet yield di erent result note experi cluster center recal recal tabl result rst set document algorithm cluster center recal recal tabl result rst set document algorithm result tabl algorithm algorithm exist ect perform classi cation perform ect algorithm present algorithm classifi web document hyperlink document base de nition di erent type node identi ed partit hypergraph set document methodolog identifi repres subset document experi show algorithm ectiv classi cation refer hypertext markup languag speci cation html work group buneman semistructur data principl system cheeseman stutz bayesian classi cation theori result advanc knowledg discoveri data page cost salzberg weight nearest neighbor algorithm learn symbol featur machin duda hart pattern classi cation scene analysi john wilei jack edmond greedi algorithm mathemat david jon prabhakar raghavan infer web commun link topolog acm confer georg kumar text categor weight adjust neighbor classi cation technic report pearl reason intellig network plausibl infer jon kleinberg authorit sourc hyperlink environ proc symposium discret daphn koller sahami hierarch classifi document word proc int conf machin low similar metric learn classi er neural maarek ben shaul automat organ bookmark content proc intern world wide web introduct conjug gradient method pain ron mark peter david gi ord hierarch network search eng exploit hypertext cluster seventh acm confer vasant le miller mobil intellig agent document classi cation machin learn approach fourteenth european meet cybernet system research articl process latex macro packag llnc style 
ML/336 6 autonom evolut gait soni robot fujita group laboratori soni corpor japan yamamoto abstract trend robot leg robot issu leg robot develop gait typic gait develop manual paper report result autonom evolut dynam gait soni robot fit determin robot digit camera infrar sensor system evolv faster dynam gait manual develop introduct paper present implement autonom evolutionari algorithm develop locomot gait process handl robot onboard comput individu evalu robot sensor implement successfulli evolv pace gait robot pace gait significantli outperform previou gait addit achiev desir goal automat develop gait result show ea real robot evolv behavior method automat creat locomot control directli applic robot set standard definit interfac entertain robot architectur featur style flexibl flexibl user reconfigur robot current implement compon build gregori visit brandei trademark soni corpor robot robot current robot develop crawl gait pattern easi build robot mechan sensor base method adapt exist locomot control robot gener gait robot robot freedom show autonom evolut behavior physic robot practic argument evolutionari robot attack evolut real robot evolut methodolog develop simul evolut shown success set acknowledg case methodolog work high accuraci occur desir evolv physic robot problem real robot mainten power suppli tether khepera power mainten larg problem evolutionari run measur hour dai interchang part easili replac malfunct leg work time problem evalu quickli evolv individu gener hour evalu unreason evolut approxim dai evalu minut physic robot accuraci readili simul evalu time short previou autonom evolut actual robot evolv behavior compar complex robot evolv control task requir control actuat exampl autonom evolv behavior backward stop behavior wheel robot home navig khepera behavior khepera behavior difficult implement hand difficult evolv simul behavior successfulli transfer simul physic robot test problem develop locomot control dynam gait dynam gait sensit behavior build simul difficult hand develop paramet difficult gait evolv experi outperform gait result show physic robot evolut robot behavior evolv rest document organ section review relat work evolv gait leg robot section descript robot locomot modul section consist descript evolutionari algorithm evolut section describ setup experi robot sensor section present result experi discuss result section section work relat work develop locomot gait leg robot problem studi year research issu popular research group section limit review relat work evolv locomot control physic robot describ gait robot gait robot capabl common pace gait crawl gait consist move leg turn leg support static gait robot center graviti insid triangl support leg contrast pace dynam gait safe area robot center graviti gait consist match diagon leg move pace gait leg side robot move figur show leg posit pace complet swing cycl leg leg ground front left front rear left rear crawl pace time figur gait research evolv gait leg robot simul evolut neural control evolv simul network transfer physic robot produc smooth walk behavior simul built ii robot genet algorithm evolv locomot control unlik exampl robot actuat control leg control consist binari string bit appli voltag wire robot neural control evolv simul obstacl avoid locomot gait perform physic robot similar simul walk gait observ simul physic robot robot simul develop evolv walk behavior exampl control evolv simul produc work actual robot similar work real robot evolut ann evolv control tripod gait robot stage evolut evolv oscil evolv walk behavior fit determin cellular encod interact evolv control input fit individu assist evolutionari process stage evolut evolutionari approach develop gait lab creat gait hand develop crawl gait gait crawl pace gait develop good time move robot backward gait develop differ work evolut complet autonom evolv dynam gait autonom evalu involv coordin sensor determin locat measur distanc challeng problem compar perform evolv control develop hand find evolv control joint inform locomot modul sensor feedback evolv control recov detect occurr robot platform locomot modul robot experi soni robot head leg freedom dof tail give total dof bodi length includ head approxim length leg shoulder onboard head stereo infrar touch sensor touch sensor bottom leg addit cpu bodi gyroscop acceleromet detail descript robot movement leg control locomot modul modul control robot gait set paramet updat joint angl leg posit swing phase modul recov robot fall movement robot develop variabl gain control algorithm vari gain schedul roll yaw pitch figur robot coordin system paramet paramet specifi minimum gain maximum fix maximum specifi swing cycl start reduc gain durat gain reduc maximum minimum back maximum sin gain phase leg phase start ffi swing forward start swing back ffi end ffi pace leg side bodi leg phase leg opposit side bodi ffi phase twenti paramet defin gait locomot modul tabl list gene individu evolv evolutionari algorithm paramet posit orient swing path rate swing amplitud oscil bodi locat gain vari swing cycl leg evolutionari algorithm evolutionari algorithm ea run onboard robot consist initi phase evolut popul paramet phase ea chosen base valu research experi section describ select reproduct phase ea evalu section tabl paramet list gait paramet unit initi rang pace bodi center mm bodi center mm bodi pitch leg mm front leg mm rear leg mm step length na swing height mm swing time ms swing na switch time ms ampl bodi mm ampl bodi mm ampl bodi mm ampl yaw ampl pitch ampl roll min gain na shift length initi popul creat uniform distribut search rang tabl list twenti paramet gene initi search rang initi rang determin experi hand develop gait individu creat evalu dynam gait paramet configur result robot fall gener initi popul set paramet initi popul robot fall replac randomli gener individu individu initi popul evolut begin tournament select select individu parent individu replac algorithm decid perform recombin mutat number individu randomli select tournament individu randomli select mutat individu randomli select higher individu lowest fit replac offspr mutat recombin variat equal probabl select mutat recombin recombin individu parent creat child individu gene child ff ith gene child ith gene parent ff random number rang mutat parent individu perturb valu small amount gener child individu random number distribut determin number gene mutat gene mutat select randomli select gene mutat ffi mutat ffi uniform random rang valu set paramet initi search rang problem experienc initi experi individu receiv significantli higher fit inaccur measur distanc infrar distanc sensor result pull search poor paramet reduc problem individu number time store ag increment time individu parent recombin mutat individu reach ag ag reset method desir result experi set paramet move robot quickli straight line section describ experi individu evalu evolut place insid pen figur end pen strip color cloth mark center end camera robot turn center color strip cloth robot measur distanc infrar sensor proce locomot fix amount time robot stop end time encount wall robot pan head find color strip measur stop distanc distanc robot score test locomot paramet calcul averag speed trial individu fit averag locomot trial center robot color strip dedic lsi chip color detect tabl detect color rang pixel posit color figur pictur environ strip robot turn gait turn place robot turn fix direct desir color detect detect robot find averag horizont locat pixel desir color convert angl camera cover angl ffi averag locat fall ffi center period robot center head pan similar center behavior robot bodi remain fix head turn color strip detect robot calcul offset current head angl rotat head face color strip distanc measur infrar sensor locat robot head infrar sensor return convert distanc creat function robot pen fix distanc color strip averag read test set paramet begin robot test distanc color strip determin reliabl rang robot maximum reliabl rang crawl gait move closer closer infrar sensor read robot wall give slow stop prevent robot run wall infrar sensor read robot bodi move normal posit start distanc determin averag consecut infrar sensor read robot target color strip move forward crawl robot move color strip figur robot pen amount time set locomot paramet stop robot fallen onboard current individu score robot individu robot trial end successfulli robot pan head find color strip stop distanc determin averag consecut infrar sensor read simplifi optim veloc straight score trial product veloc straight score averag veloc robot trial straight function angl robot forward direct direct target color distanc target figur calcul straight convert measur offset function straight normal account robot distanc color strip robot fix orient larger robot closer color strip function defin score start stop stop start stop start stop time stop stop ffi function constant maximum minimum measur distanc tabl list valu stop valu stop robot find color strip assum robot gait turn sharpli pan head face color strip case receiv score score receiv ffi individu select averag score trial tabl sampl valu stop stop stop ffi ffi ffi ffi ffi ffi ffi ffi ffi method run experi evolv gait evolv pace gait differ experi time set leg swing gait set leg swing ffi shift leg pace gait set leg side robot move left side ffi shift result evolv gait popul size ran gener random individu creat make initi popul individu initi popul individu move move backward move awkward individu tendenc walk curv end evolut individu propel robot smoothli straight individu fit score move minut trial figur graph plot result evolv gait figur graph show result evolut pace gait initi individu gait stabl gait randomli gener individu creat initi popul paramet initi popul coupl good individu move quickli individu fit gait larg varianc perform trial locomot paramet averag score trial gener averag figur gait result allevi maintain individu ag reproduct gener evolut individu move figur sequenc imag evolv pace gait interv frame gener averag figur pace gait result experi gener approxim hour evolv paramet individu pace gait list tabl discuss implement evolutionari algorithm enter fit score addit distanc move robot object measur aesthet adjust individu fit find paramet dynam gait errat notabl earli stage evolut set paramet perform evalu poorli figur pace gait figur gait averag fit multipl trial result evolut dynam gait consist perform find reason averag perform trial autonom evolut hand gait pace gait evolv autonom evalu fit evalu metric metric distanc fix amount time evalu metric inaccur individu improv individu start trial middl pen left approxim wall good individu easili allot time happen row score trial good individu frequent low score result low upper limit evolv chang variabl time trial score averag speed upper limit factor limit maximum evolv speed variat start angl robot initi plan run pace experi gener gener improv differ fit result start angl robot center color strip robot spent switch gait paramet current individu turn slightli robot penal stop larg ran straight result differ fit top individu matter luck select pressur individu popul improv experi produc pace gait faster gait true manual evalu automat evalu reason hardwar design pace gait robot shift bodi assist lift leg ground gait leg side bodi clearanc ground move leg difficult robot robot twist torso twist bodi lift leg ground achiev good perform pace evolv gait dynam gait evolv drag forward move leg ground run experi robot angl bodi forward rest front leg run rear leg drag ground individu crawl front rear leg slide ground dynam gait stabl leg eventu fall leg stabil move leg ground switch robot time fall leg evolv earli individu move fast rest leg individu lean side tend turn move receiv larg penalti stop consist rest leg side bodi remov drag robot move straight line graph pace paramet initi random popul good hand tailor control case run individu initi popul show period minut cover fit move smoothli straight bounc frequent turn averag fit popul indic perform evolv paramet robust robust end individu offspr tend similar success individu robust mutat success offspr paramet flourish paramet sensit mutat offspr success di individu sensit environ leg calibr voltag individu perform differ typic consist frequent place move curv ad sensor feedback reduc affect find suggest individu special environ evolv gener gener control individu evalu environ evolv dynam gait hard robot month develop test evolutionari algorithm frequent mainten neck repair cpu board exchang wire leg replac time dozen rubber feet simul easili construct problem requir fine control simul option paper present work autonom evolut dynam gait evolv vector paramet locomot modul evolutionari algorithm run onboard robot robot sensor fit individu evalu assist algorithm successfulli evolv pace gait robot tabl show fit valu individu initi popul individu found actual speed individu pace gait move robot approxim significantli hand develop gait unsuccess develop pace gait tabl summari result initi pop fit fit speed pace previou work evolut real robot simpl robot evolv control task minim simul robot built similar control successfulli evolv complex robot dof evolv sensit behavior locomot sensor achiev result control result show feasibl evolv behavior real robot futur work plan evolv behavior simul model lowlevel behavior evolv real robot acknowledg author ronald jun member soni group refer simpl climb run proc intern confer robot pp francesco evolut home navig real mobil robot ieee transact man stefano competit theori practic anim animat mit press open architectur robot entertain del marina proc intl conf autonom pp hiroaki develop autonom robot robot entertain autonom john randal qualit dynam analysi evolv locomot control herbert anim animat pp applic evolv locomot control robot robot autonom freder cellular encod interact evolutionari robot technic report sussex develop honda humanoid robot proc intl conf robot pp nick minim simul evolutionari robot phd school cognit comput sussex andrew alan genet program approach construct neural network control walk robot proc conf robot pp maja dave challeng evolv control physic robot evolutionari gari david ingo learn gait proc intl conf advanc pp evolutionari robot intern laboratori microprocessor swiss feder institut technolog luc emerg function robot agent evolut proc fourth workshop artifici pp embodi evolut develop humanoid robot antagonist driven joint dof trunk intl conf intellig robot pp 
HCI/1288 4 ubiquit web inform agent gerd bernd christian wolff institut comput scienc universitat koblenz abstract paper overview ai method techniqu develop build ubiquit web inform system method area machin logic knowledg system discuss context prototyp inform system mia mia web inform system mobil equip pda cellular phone gp devic cellular wap phone captur main issu ubiquit locat anytim inform access pda technolog introduct biggest chaotic unstructur sourc inform make amount inform ubiquit comput daili life great challeng hardwar issu wireless ubiquit solv wearabl comput integr pda telecommun major problem intellig inform extract www overwhelm mobil user document found offer short piec inform interest good inform system present address nearbi restaur retriev onlin web match prefer cuisin condit hold awar user react directli user present inform found user inform extract onlin web requir special prepar web page page special tag addit develop method artif intellig machin logic knowledg system paper give overview ai method techniqu build ubiquit web inform system prototyp test method system mobil inform agent main intent paper show implement ubiquit web inform system ai techniqu suitabl solv emerg problem mia system mia base inform system retriev short fact anytim fast queri time monitor posit mobil user autonom updat subject search occur user travel search interest section organ section describ briefli basic compon mia detail introduct ai method spider agent section conclud section mia session architectur mia system separ basic compon mobil server system mobil mobil comput agent local agent user agent matchmak mobil agent server init request agent commun fig system architectur user equip estim geograph posit commun wirelessli server locat internet mobil phone develop mobil agent mobil phone wap support system wai inform current posit read user add posit manual autom geograph posit gp gp work close work place lot build inconveni user interest system popul area bad percept gp signal eleg overcom problem mobil phone connect inform determin posit user mia prototyp work gp support manual posit inform geograph coordin deliv gp devic resolv local agent track user estim nearbi citi gatewai core base retriev system http protocol commun html wml page mobil agent core inform system consist interact agent current prototyp work agent user agent model local agent spider agent web search find inform relev current geograph posit regard interest user distinguish type inform search web topic extract topic web page inform relat user interest current citi restaur extract inform topic descript descript topic extract present profil web site structur profil configur gener specif mia method defin user profil captur hierarch structur structur plai import role spider section model topic constrain keyword list set constraint oper constraint defin topic restaur extract chosen predefin set due inform extract capabl mia prototyp support oper address extract mia user person profil consist topic figur show profil titl trip consist topic topic assign extract address user chang profil inform retriev user agent propag current user profil agent ai method intellig web inform retriev section overview artifici intellig base techniqu spider agent classifi extract inform web spider step inform retriev process find web page inform special topic contrast exist search approach search relev web page onlin call approach index web document due estim word frequenc build index determin relev web document relat page approach interpret www direct graph web page node hyperlink vertic label url text assum keyword start page found relev web page due keyword start start page find path node vertic label url vertex label path keyword keyword occur vertic label basic idea improv search site inform restaur follow path label edg link occur path introduc paramet keyword defin context radiu keyword similar cut search path depart occurr keyword occur kn label follow path paramet kn call defin approach set entri point set keyword start return link iff exist cycl free path node vertic connect wm label wm label kn kn wm label find path modifi search strategi mention keyword distanc paramet cost function similar standard inform search algorithm path link decid link set link nearest goal chose link path minim keyword distanc spider agent search keyword link follow path occur time link agent prefer follow web page classif ad text classif compon improv effect effici system web page provid spider algorithm address filter irrelev web speed system significantli effici inform extract compon inform facilit task extract inform web page inform extract agent extract address web page varieti method method address inform posit page addit inform inform extract compon make method classifi high confid qualiti web inform extract agent method method fail web page inform extract agent method simpli abandon page artifici neural network text classif network architectur chosen fulli connect network input hidden output node map web page input node select input node web page input node output node repres classif page page train output node set output node set recognit document put class output node higher activ neural network train offlin preclassifi page page authent web page gather spider algorithm network train error longer declin independ set page page train set classifi correctli excel classif result partli due distribut web web page gather spider agent address posit recognit rate word give web page chosen select word word web page train set inform word distinguish page page address machin learn inform extract extract inform web page understand involv semant text analyz find relev inform recogn underli syntact linguist motiv attempt semant syntact fail web document make special text format annot string document recogn extract relev inform syntax base approach automat learn extract procedur similar approach machin learn techniqu automat wrapper construct extract assum special text part delimit mark begin end relev inform extract kei idea learn algorithm set extract web address set address list collect surround text part compon address learn gener pattern anchor combin learn pattern addit constraint result learn wrapper intend inform extract task illustr gener idea learn wrapper user read web page list relat instanc relat inform found list shown web page user determin instanc list instanc posit set exampl learn task find definit relat address exampl list address instanc relat address mean learn extract procedur extract instanc remain instanc present page gener techniqu base combin method adopt area induct logic program detail reader refer major problem confront context mia lack exampl classifi unknown web page determin learn overcom problem develop learn algorithm deriv exampl mean knowledg techniqu model structur knowledg address logic kr languag queri knowledg base deriv exampl input modifi learner learn wrapper unknown page call approach learn meta exampl automat construct address techniqu show result mia session demonstr gener behavior function mia small imagin plan bit unsur citi visit journei interest address restaur offer chine cultur institut cinema home mia interest standard web browser interfac web browser instal pda start select mia applic fig pda result store pda activ profil mobil devic contact mia server cellular phone internet connect transmit current geograph posit preselect time interv user interact initi profil server retriev posit matchmak request mobil user extend matchmak decid initi user agent spider agent fulfil request agent activ inform user agent track locat search interest search interest defin chang move locat report user commun local agent local agent resolv geograph coordin nearest citi user agent inform activ spider agent perceiv spider agent start web page topic interest current citi drive nice town decid stai time time mia found select mia applic click button result user agent queri inform found activ spider agent result agent constantli listen request immedi task answer incom request result found return directli user agent final mobil user pda logic program base agent technolog kei issu mia architectur logic program languag agent agent commun languag techniqu agent control queri agent flexibl dynam initi assign agent handl matchmak special agent job match inform search inform provid agent track agent user task extend matchmak concept check simpl subsumpt techniqu incom request split handl activ agent techniqu spider agent prevent system agent search topic function agent mobil agent matchmak user agent spider agent logic programm lookup spider agent fig agent commun logic program program describ search web knowledg acquir agent ad uniform knowledg interfac base interfac base common inform exchang level type agent agent necessarili rule agent program activ agent receiv request actual basic task interrupt incom queri process rule agent program fire common defin uniform predic describ state inform acquir agent trust special servic builtin logic queri interest point agent add knowledg form rule agent logic program agent fix set commun rule similar kqml execut commun rule result consult matchmak agent type activ user userid agent commun directli activ agent matchmak initi call agent send initi command initi agent major advantag interrupt logic program base agent possibl provid anytim inform access agent ist found wait complet task interrupt basic task queri interfac base retriev result found send back work essenti mobil inform provid user anytim fast inform user send format retriev request mia system queri send matchmak user agent ua user incom request interpret proof ua proof queri appli rule result execut commun rule result matchmak activ agent type spider user small assum spider agent activ user search address chine restaur interfac base consist address found queri result comput knowledg final ua fulfil proof oblig send answer mobil agent futur work show detail major ai logic machin learn system combin build ubiquit inform system gener agent architectur base interrupt logic program techniqu implement anytim info access present section gave short introduct exist palm gp system cellular phone build mobil comput unit provid locat anytim inform access wireless commun point problem gp posit estim inconveni prototyp setup consist separ devic lead futur plan import point capabl mia retriev inform unmodifi web page limit special search web inform system access upto date onlin inform describ ubiquit system mia citi inform system stationari user pc home futur work extend pure keyword base web search algorithm deduct reason process deriv relat term keyword ontolog constraint keyword list combin deduct ontolog base enrich retriev process domain specif knowledg topic restaur deduct resolv queri restaur french italian offlin train web page classif train network creat adapt agent onlin feedback extract agent adapt network web result extract teach input pattern neural network start work eleg overcom problem relat gp base local cellular phone inform cell connect cell area serv fix basi station telephon network inform basi station phone connect geograph posit basi determin posit user fairli accur high resolut gp harm high resolut gp resolut price map significantli lower resolut gp moment switch html xml serv html page format serv page standard document format acknowledg work present paper fund land ur land ur und project intellig web inform agent refer califf relat learn techniqu natur languag inform extract phd texa august cover thoma element inform theori mayfield softwar chapter kqml agent commun languag mit freitag machin learn inform extract inform domain phd comput scienc carnegi mellon novemb thoma unifi approach html wrapper learn technic institut fur knight multidisciplinari survei acm comput march kushmerick wrapper induct inform extract phd lloyd foundat logic program muggleton raedt induct logic theori method journal logic plotkin automat method induct infer phd goebel comput logic approach oxford rumelhart mcclelland parallel distribut explor microstructur volum foundat mit thoma base learn inform extract workshop machin learn inform juli nation american confer artif intellig thoma logic program intellig web search intellig inform special methodolog intellig inform system editor multiag system mit wiederhold mediat architectur futur inform system ieee page march 
ML/637 6 learn languag collect case tune paramet yasubumi sakakibara klau jantk steffen lang institut social inform fujitsu laboratori japan htwk leipzig fachbereich germani abstract investig problem learn formal languag reason learn area artifici intellig formal framework learn languag recent develop induct infer manner show index class languag finit decid class languag includ class regular languag learnabl fix similar posit neg exampl present framework learn learn algorithm learn similar avoid trivial encod carefulli examin extent similar measur learn learn paramet similar show index class languag finit problem decid learnabl impli languag learnabl collect case learn paramet similar measur introduct reason deem import technolog widen bottleneck knowledg acquisit ai knowledg repres form case similar measur form rule enorm amount approach applic learn investig natur design learn procedur main task learn collect good case store case base describ knowledg classifi unknown exampl learn algorithm construct explicit gener exampl learn algorithm deriv investig power limit learn algorithm work manner introduc appli formal languag formal framework learn recent develop induct infer manner induct infer theori learn incomplet inform learn algorithm architectur behavior motiv artifici intellig research relev theori induct infer ai applic dealt induct infer exampl counterexampl learn formal languag studi area induct infer remark amount work induct infer index class formal languag index class effect formal languag admit uniform procedur decid membership formal languag accept list investig consid index instanc chosen languag target object tailor knowledg process kei reason prototyp case occur constitu list describ area gain insight fundament phenomena learn research focuss effect learn similar concept tune defin paramet section show index class languag finit decid show assumpt underli similar measur fix learn learn class languag includ class regular languag posit neg exampl shown interest class formal languag learnabl assumpt section framework learn learn algorithm learn similar avoid trivial encod carefulli examin extent similar measur learn interest import method learn similar measur adopt weight scheme case weight nearest neighbor algorithm accomplish weight similar reliabl case larger make closer word domain learn paramet weight similar show index class languag finit problem decid learnabl section appli theoret result learn list learn formal languag give formal definit learn formal languag introduc show gener formal languag framework index class languag finit problem decid learnabl fix similar measur definit learn finit alphabet finit set denot set string denot set string languag set string languag denot complement denot empti string similar measur oe defin similar string comput function real interv case base cb finit subset call case cb posit case neg case standard semant compet consid resolv conflict classif adopt standard semant standard languag repres similar measur oe finit case base cb defin restrict posit case target languag neg case complement languag case base cb describ finit subset natur requir learn induct infer manner definit learn similar measur priori fix target class languag problem learn problem collect case base learn similar measur learn denot set posit natur number target languag learn inform infinit sequenc element fs ing fs fs element form call posit element form call neg denot set inform inform natur number denot initi segment length denot segment nth element languag call consist posit exampl neg exampl learn algorithm input larger larger initi segment inform output sequenc converg limit element natur number natur number abbrevi lim class languag learnabl inform learn algorithm similar measur oe exist case base cb lim case base cb finit subset set label word write cb fin formal languag prove index class languag finit problem decid quot lemma proof straightforward theorem lemma index class languag singleton languag infinit languag similar measur oe repres languag case base consist posit case order prove encod techniqu develop effect word repres languag relat word languag artifici effect languag index class languag membership uniformli decid wrt total function decid string theorem index class languag finit problem decid similar measur oe languag repres oe finit case base cb posit neg cb fin proof defin similar measur oe satisfi desir fc denot set finit languag belong pc denot set infinit languag fc pc languag finit problem effect languag effect languag order repres finit languag fc result lemma requir counterexampl denot complement construct word word member word belong assum word construct proce step collect word denot minimum set ith step defin fu fu fu similar measur oe defin standard definit addit ae ae ae membership problem languag uniformli effect defin oe effect comput infinit languag case base finit fw case base wrt oe case repres finit languag neg case complement satisfi requir posit case target languag neg case complement languag corollari class languag proof finit problem languag decid result theorem result gener neg answer learnabl formal languag theorem class finit languag learnabl inform proof assum learn algorithm work wrt similar measur oe deriv contradict start observ learn inform cb easili defin inform actual case base infinit time fail infer inform notic cb posit case behavior fed inform finit languag fw initi segment learn cb cb cb string cb cb counterexampl case base cb belong meet cb obtain contradict assum string satisfi requir finit sequenc defin initi segment inform languag fvg cb due choic conclud cb element cb fail learn inform initi segment fed add posit case actual case base cb accumul case base repres oe assum inform finit languag initi segment neg exampl past point cb cb obtain subsequ add neg case actual case base cb produc case base cb oe cb fail infer finit languag fug contradict prove theorem corollari class regular languag learnabl inform proof clear class regular languag finit languag languag learn similar measur tune paramet consid similar measur fix target class languag learn develop case base fix similar measur restrict learn similar measur learn target languag class class similar measur oe finit case base cb account class comput similar problem reduc defin learn similar measur interest view point learn carefulli class similar measur consid introduc paramet similar measur learn interest class similar measur adopt weight scheme case weight nearest neighbor algorithm scheme base idea store case reliabl classifi accomplish weight similar reliabl case larger make closer word domain class languag assum fix base measur oe weight similar measur oe base oe defin ff oe ff comput real valu number construct weight similar measur oe defin languag finit case base cb ff simplifi concept weight similar measur oe base oe ff oe cb oe cb denot class weight similar measur oe base oe weight similar measur simul forget case simpli set weight ff class languag class similar measur learnabl inform wrt learn algorithm exist case base cb similar measur oe oe oe cb lim problem learn similar measur problem tune paramet ff ff learnabl wrt weight similar measur show index class languag finit problem decid learnabl inform wrt class weight similar formal languag shown theorem fix similar measur theorem index class languag finit problem decid learnabl inform wrt class weight similar measur proof denot target class languag learn order repres finit languag complement finit languag pc denot set infinit languag fc denot set finit languag pc finit languag fc denot result element odd indic pc fc natur number achiev infinit sequenc repres word languag pc complement finit languag fc construct effect member belong construct complet similar skip short base similar measur oe defin oe ae oe ae note infinit sequenc repres infinit sequenc repres weight similar measur oe defin ff oe case paramet ff construct learn algorithm assum consist induct infer algorithm base identifi limit produc output indic induct infer algorithm realiz identif principl wrt behav requir construct simul base inform target languag base denot index gener base fed sake write base shorthand languag guess base algorithm input figur assum cb flag algorithm begin read dx dx finit languag ff sy dy infinit languag sy ff sy finit languag sy ff sy ff sy end fig learn algorithm formal languag note algorithm effect comput halt produc base assum output indic decid base finit start observ immedi definit oe definit observ base describ infinit languag flag base oe cb denot actual case base oe actual weight similar measur guess cb repres oe definit oe obtain ff exampl cb ff sy base oe order show behav requir distinguish case case infinit languag base infer base base base cb repres converg final case base cb cb weight similar measur oe oe construct flag obtain observ infinit repres case converg final case base cb cb similar measur oe oe flag case finit languag inform base infer base base appli argument conclud converg case base cb repres case complement base oe denot weight similar measur converg definit cb fw cb cb impli assum cb ae string cb order show suffic show cb due definit oe oe choic definit prevent store posit case actual case base cb requir base base flag fulfil base infinit languag observ made conclud base oe base consist induct infer algorithm base base impli base base base oe cb oe due choic oe oe cb class regular languag languag learnabl inform wrt class weight similar measur gener learn algorithm figur full set exampl present point requir collect case base start empti case base correspond condit definit type learn call increment usual increment learn defin induct infer infer algorithm back histori inform present learn process formal definit increment learn chang condit definit oe dn oe cb empti case base oe assum initi similar measur learn algorithm figur simul base algorithm base index class languag finit problem decid exist increment consist induct infer algorithm increment learnabl inform wrt class weight similar measur increment learn import reason learn intend model human expert learn behavior consid learn process long time period mean updat theori process recent recogn histori theori develop adequ learn model essenti increment applic learn list introduc class formal appli theoret result learn list introduc studi effect learn languag pac learn framework learnabl studi class list simpl class defin formal languag base list class languag defin subclass regular languag definit list simpl defin relat string uw list list pair dn string label list defin denot string label equal index item ff true ff aa true ff ccc true fig diagram list list rule list diagram figur cg pair list theorem conclud class languag accept list wrt fix similar measur class finit languag repres result list deriv theorem class list learnabl inform gain class tunabl similar measur base defin simpl natur similar measur oe cont string oe cont list weight similar measur base oe cont class list list repres weight similar measur oe base oe cont finit case base oe defin ff oe cont real ff proof dn list repres case base cb weight similar measur oe base oe cont defin cb dn oe cont oe cont string length clear shown class list learnabl increment appli class similar measur defin theorem reflect order repres case case base adopt idea obtain statement class list learnabl inform wrt cont proof properli sequenc expon kn defin list kn kn assum inform initi segment continu sequenc case form exce kn sequenc learn suffici larg put current case base stabil learn process inform complet provid inform case occur learn algorithm investig necessarili fail kei reason normal form result list develop gener suffici list node kn kn similar normal form result built cont learn algorithm drop case collect case base case present fail natur similar measur introduc learnabl seemingli unnatur approach invok prove theorem properli greater learn power yield result theorem class list learnabl inform wrt class weight similar measur proof languag repres list invok decid finit regular languag proof straightforward theorem determinist approach list shown learnabl pac learn model algorithm present theorem naiv simul identif strategi effici learn learn essenti base collect case tune similar measur appropri attempt natur formal area tailor reason result exhibit enorm power learn similar class similar measur account approach fix similar concept assum coincid result learn function methodolog discov import similar concept learn seemingli natur class similar measur admit learnabl concept nice illustr limit reason autom learn stress import approach firmli base theoret comput scienc result cognit scienc idea formal approach reader consult coupl similar result respect detail inspect construct proof exhibit import distinguish case repres target languag learn fashion question formal cognit scienc concept prototyp acknowledg work author partial support deutsch forschungsgemeinschaft worth mention origin present investig fujitsu laboratori author opportun collabor refer david denni marc albert learn algorithm machin dana angluin induct infer formal languag posit data acm comput dana angluin carl smith induct infer theori method inform scott cost steven salzberg weight nearest neighbor algorithm learn symbol featur machin mark gold languag identif limit inform john hopcroft jeffrei ullman introduct automata comput klau jantk learn induct infer workshop comput learn theori page acm klau jantk type increment learn work aaai spring symposium train issu increment page stanford klau jantk steffen lang learn pattern languag workshop algorithm learn theori lectur note artifici intellig page ronald rivest learn list machin yasubumi sakakibara siromonei model learn set string workshop comput learn theori page acm articl process macro packag llnc style 
HCI/1794 4 accept public distribut cite circul design evalu battlefield visual virtual environ deborah hix edward ii joseph mike mcgee jim durbin toni king virginia polytechn institut state va naval research dc virtual prototyp simul va abstract power comput hardwar render system concentr creat visual rich perceptu realist virtual environ applic compar effort expend user interact compon ve ve user interfac poorli design rare evalu user usabl engin newli emerg facet ve design usabl evalu ve practic lag paper present structur approach design evalu user interact compon ve approach consist iter expert heurist form usabl summ evalu paper describ approach ve applic battlefield describ seri design present evid approach strategi assess iter improv user interact design ve user user user usabl usabl virtual virtual expert heurist form evalu introduct relat work power comput hardwar render user interact compon ve applic poorli design rare evalu user vast major ve research design effort develop visual qualiti render effici visual compel ve difficult user ve make good entertain usabl problem prevent effici solv problem usabl engin newli emerg facet ve design evalu ve design develop awar tradit interfac usabl research begin appli expand method ve effort report design usabl evalu ve practic lag report work evalu ve investig aspect navig ve present framework evalu viewpoint motion travel framework support methodolog evalu ve travel techniqu appropri match travel techniqu virtual applic qualiti identifi import spatial inform user comfort author acknowledg factor system greater impact qualiti factor perform travel techniqu select evalu methodolog intend generaliz varieti ve discuss usabl engin method shape iter develop ve design educ student concept newton law physic goal design process develop usabl educ virtual world author appli usabl evalu identifi refin earli system weak learn usabl potenti user school expert field particip form result improv final ve user interact research report limit element usabl evalu includ studi haptic interfac investig spatial input devic describ basic method evalu gener usabl compon ve effort provid insight usabl issu specif ve provid suffici breadth complex ve design assess framework usabl characterist structur support usabl engin ve present methodolog approach design assess ve user interfac emploi refin ve usabl space framework evalu battlefield visual ve section section personnel naval research laboratori virtual realiti lab develop ve battlefield call dragon implement workbench workbench natur metaphor visual interact accept public distribut cite circul figur screen shot dragon battlefield visual virtual environ scene familiar tabletop environ applic user collabor excel candid workbench research collabor research virginia empir studi import usabl paramet effect ve user interfac dragon section discuss battlefield visual describ dragon battlefield visual ve section discuss import usabl evalu method profit appli expert heurist form summ evalu section present methodolog approach appli expert heurist form evalu method dragon design section describ discuss dragon iter result approach conclud section discuss lesson learn includ evid structur approach strategi assess iter improv user interact design idea futur summ evalu dragon battlefield visual virtual environ battlefield visual tradit battlefield visual accomplish place paper map sheet intellig report arriv technician pencil mark inform command draw plan direct battlefield situat map present visual map overlai hour updat histor paper oper perform liter box fill sand shape replic terrain command move small physic replica battlefield object direct battlefield situat modern battlefield produc inform method inadequ effect visual accept public distribut cite circul workbench displai observ manag inform share technician command visual inform includ terrain entiti repres neutral repres featur obstruct kei map point dragon receiv electron intellig feed provid inform entiti current includ damag user navig observ map entiti angl queri manipul entiti design user interact dragon earli dragon produc assess gener interact method interact hand gestur speech interest possibl ve found speech recognit immatur battlefield visual found pass user work user hand approxim size found handheld easili hand user work user base dragon modifi game remov base place posit sensor insid posit orient track rel emitt locat front center workbench laser pointer laser beam user point ve map beam intersect terrain highlight marker earli usabl evalu dragon demonstr prototyp system militari object proof concept battlefield visual feedback civilian militari user found dragon accur current visual effect effici tradit method pencil success began intens usabl evalu iter dragon user interfac usabl evalu method evalu kei compon develop interact import applic complex innov ve kind usabl evalu expert heurist form summ evalu perform type extens dragon develop plan type section expert heurist evalu type analyt evalu expert user interact design assess user interfac determin usabl design guidelin violat support base expert make recommend improv design case challeng guidelin specif ve user interfac heurist evalu expert user interact repres user directli involv type usabl evalu typic effect expert develop user interact design evalu author involv develop expert heurist evalu section form evalu type observ assess user begin earliest phase user interact design continu entir life cycl form evalu produc qualit quantit result form evalu iter quantifi assess improv user interact design import point note form evalu shown figur qualit quantit data collect repres user perform task scenario develop impress usabl evalu warm process collect data contrari experienc usabl evalu collect larg volum qualit data quantit data figur form evalu process qualit data typic form critic incid critic incid occur user perform task event develop user task scenario repres user perform scenario protocol evalu collect qualit quantit usabl data suggest improv user interact design refin user task scenario repeat accept public distribut cite circul posit user task perform user satisfact interfac event affect user perform satisfact impact usabl typic critic incid problem user encount unabl complet task section describ major design iter occur result collect hundr critic incid form evalu studi quantit data gener long number error user perform task scenario data compar baselin metric quantit data gener problem qualit data occur collect type data kei part form evalu process larg critic incid collect quantit data summ evalu empir assess user interact design comparison interact design perform user task summ evalu typic perform version interact yield primarili quantit result summ evalu statist compar user perform interact determin defin advanc summ evalu dragon plan guess interact design substanti refut short cycl heurist form quickli assess numer aspect characterist interact design design assess dragon ve user perform numer cycl heurist form short minut bad hour evolut essenti design detail round evalu discuss achiev optim design dragon user heurist form plan summ studi applic design evalu method focu navig earli demonstr observ navig user manipul viewpoint move place place virtual world map battlefield affect user task user successfulli navig move virtual user task involv specif object group simpli perform user queri object user navig virtual world object perform user task analysi heurist form studi corrobor task analysi expect task import observ studi reveal gener task perform user type applic includ object object object queri object aggreg user task focu futur research perform expert form usabl guess assumpt user task methodolog basic dragon applic instrument modifi heurist form usabl evalu perform extens evalu user cycl evalu singl evalu session uncov design problem pointless user attempt perform scenario design iter base begin cycl evalu major cycl iter base task analysi earli creat set scenario benchmark user carefulli consid coverag specif issu relat navig task exploit egocentr move navig metaphor exploit move navig metaphor section scenario navig task virtual map world scenario serv prime explor search design evalu rate control versu posit control virtual world scenario present user evalu session expert heurist evalu expert heurist user interact design expert work collect assess evolv user interact design dragon earliest heurist expert follow specif user task scenario engag simpli user interfac expert knew dragon battlefield visual ve explor kind task import user dragon heurist evalu person typic hold gener decid explor applic expert observ comment discuss occur session mention author paper expert assess current design assess discuss guid larg knowledg interact design framework usabl characterist ve accept public distribut cite circul discuss section framework provid structur mean evalu wander random provid guidanc make modif improv discov violat design guidelin found dragon interact design major design problem uncov expert heurist evalu includ poor map navig task miss function terrain problem damp map movement graphic textual feedback user current navig task problem address discuss section cycl expert heurist evalu reveal remedi design flaw move form evalu form evalu form evalu formal protocol welcom give overview evalu explain workbench dragon applic care explain detail dragon interact evalu user plai figur button activ navig task time user attempt determin note comment made critic incid occur user successfulli figur began perform scenario minut pass user figur button happen fill detail determin move scenario time perform set scenario rang minut hour time user perform individu task count error made task perform typic error move wrong direct navig metaphor error involv simpli maneuv map rotat persist problem map navig task button discuss section carefulli note critic relat construct comment user made design evalu ran session interact evalu record count collect qualit data expert heurist evalu session form evalu session evalu found qualiti amount data collect multipl evalu greatli outweigh cost evalu analyz quantit qualit base iter explain section iter dragon user interact design tabl summar major iter dragon user interact design approxim period descript iter visual major usabl find iter permit explain inform tabl shown row fell gener descript descript top cell tabl gener descript iter salient featur shown approxim date iter complet interact descript categori describ specif user interact design iter experi extens variant navig metaphor egocentr visual laser pointer metaphor section iter beam iter model skeleton hand beam visual user hand edg figur categori tabl show freedom tracker design iter devic descript categori defin map button exampl explain evalu result categori type evalu perform summar major strength major flaw row tabl summar user interact design modif recommend dragon programm earli implement navig navig metaphor base user interact real physic map tabl button navig task pitch map mimic motion map act stuck laser user movement direct map move direct magnitud user gestur control distanc map movement virtual world mean pan side side user make repeat pan translat map distanc length user gestur navig metaphor base concept user fly map airplan button combin navig task magnitud user gestur control veloc map user fly side map singl gestur accept public distribut cite circul virtual point modal integr navig gener descript metaphor gestur move map navig task separ discret mode mode map button approxim date june novemb januari april interact descript navig metaphor primarili rotat primarili rotat laser pointer visual laser pointer skeleton hand laser pointer laser pointer laser pointer support freedom pitch roll pitch pitch devic descript button map trigger zoom trigger pitch zoom pitch head roll left button cycl rotat zoom head rotat zoom evalu result evalu perform heurist heurist heurist form heurist form major strength iter easi good overview task navig easi navig locat singl mode easi navig locat easi switch navig task major flaw iter skeleton hand orient match user hand orient terrain difficult pan gestur parallel floor workbench screen hard travel locat map travel underneath map trigger overload freedom navig task result unintent roll cumbersom switch mode user movement slow zoom gestur parallel workbench floor recommend programm interact design support terrain damp add detect map remov abil roll add rotat coupl mode navig mode map button coupl pitch head coupl pan zoom damp tabl major iter dragon user interact design base concept briefli section version demonstr militari mention section addit expert heurist feedback demonstr kei find iter user map base observ user interact map combat initi thought visual applic requir navig metaphor dragon creat rich environ user move map experi environ visual size terrain entiti field line navig work global manipul environ conduct oper unit user capabl logic approach design dragon fly capabl found map button navig task shown tabl trigger left button press simultan produc combin pan trigger button produc combin head work user design framework usabl characterist ve section suggest possibl navig metaphor wim ultim design attempt avoid mode navig task usabl problem mode interact base person navig object locat real point reason adopt idea navig simplifi design mimic real world user simpli point locat object accept public distribut cite circul press trigger fly found expert heurist evalu singl gestur move power support complic navig task inher dragon singl gestur meant freedom control singl gestur result unintent roll user pan zoom essenti observ control versu conveni navig task activ conveni difficult physic control separ task conveni physic control easier freedom limit mode addit found user rotat move complet tank dragon rotat ad interest find show pure pure metaphor metaphor aspect depend user goal extrem navig task coupl singl button opposit design navig task separ mode user click left dragon cycl success task small textual indic displai workbench user cycl desir user simpli move task enabl push button observ cumbersom user cycl obviou achiev conveni control framework usabl characterist ve decid fourth iter dragon interact coupl pan zoom pitch head singl rotat zoom tabl fourth gener design achiev desir conveni versu control final evalu found design navig work user problem observ damp map movement great made lesson learn futur work kei find research success progress heurist form summ evalu strategi assess improv user interact design summ studi conduct applic interact design heurist form evalu situat unfortun case summ evalu compar good appl bad orang differ found interact design due fact design inher term user interfac simpli usabl design heurist form evalu prior summ confid interact design compar essenti term compar good appl good orang differ found summ comparison due differ natur user interact design se cost perform type evalu typic lowest expert heurist evalu highest summ shown figur expert heurist evalu perform prior form form studi typic longer requir larger number reveal usabl problem gener discov heurist evalu expert heurist evalu reduc cost form form studi produc interact design compar summ studi uncov differ applic figur type usabl evalu cost futur work focu summ evalu current navig design expert heurist form discov variabl affect usabl navig ve base framework usabl observ heurist form ve interact narrow numer variabl feel critic import summ navig metaphor gestur control posit hand visual present devic cave tm head track usabl evalu type low high summ evalu form evalu expert heurist evalu cost accept public distribut cite circul expect result plan studi empir determin guidelin design navig ve research produc result import improv navig recommend design navig evidenti substanti structur approach design evalu ve paper report expert heurist evalu form usabl evalu structur approach design evalu user interact compon ve approach realworld ve applic battlefield result ve empir evid effect usabl acknowledg peopl contribut dragon develop report work eddi late normal call duti dragon readi evalu brad chri made recommend develop programm includ josh bob greg josh summer dr larri rosenblum dave tate gave guidanc project dr jim linda sibert nrl dr bob virginia tech provid valuabl suggest research fund offic naval research dr helen program unit state marin corp dr continu support synergist collabor interact research virginia tech nrl past year refer methodolog evalu travel techniqu virtual virtual applic rp jl strategi behavior larg virtual proc human factor comput chi visual proc ieee visual ieee comput societi press jl engin virtual environ framework usabl submit journal virtual environ complet framework document onlin del dr evalu critic incid softwar document proc annual human factor societi ca displai remot virtual fur virtual environ advanc interfac design oxford press nf hint spatial proc acm symposium user interfac softwar technolog pp hr develop user ensur usabl product process john wilei virtual work ieee pp usabl academ press virtual environ visual interact object scientif proc ieee visual ieee comput societi pp virtual realiti virtual world ieee comput societi press rb learn educ virtual proc annual human factor ergonom societi pp san california realiti interact world proc human factor comput chi design virtual environ virtual camera control virtual comput 
Agents/151 1 augment commerc agent intellig system laboratori swedish institut comput scienc box sweden internet evolv inform space market space potenti electron auction commerci servic creat great problem major problem difficulti find relev offer problem cope multitud style user interfac marketplac problem autom routin task environ present solut problem market agent support user custom commerci site find match negoti close deal infrastructur open decentr particip agent act interest owner interact symmetr particip plai role market paper present integr sic web person assist agent user interact servic handl routin task servic adapt interest provid mechan initi highli manner softwar worldwid electron person assist introduct internet evolv inform space market space thousand electron auction commerci servic larg number type commerc site simpl advanc shopbot start search telescop inform store sell telescop vista hit good deal shopper find store sell proce compar manual lot work good offer miss due sheer exhaust problem search engin necessarili cover part shopbot made lot easier find compar offer catalog regular end result small number player control choic search choic evalu negoti mechan market strongli conflict notion free market particip particip complet control interact companion paper previou work introduc framework market sic particip market support agent infrastructur open web enabl autom shopbot particip agent act interest owner singl site control market site serv central role indic search role open particip perform particip trivial switch interact symmetr particip plai role market describ agent integr web sic smoothli combin web autom person assist agent user interact servic handl routin task servic adapt interest provid mechan initi highli manner remaind paper organ section briefli summar sic discuss relat work section present basic integr agent web section describ basic function person assist section present usag scenario section discuss implement issu section offer final background relat work sic sic market infrastructur intend complement form commun core inform describ user interact defin basic vocabulari negoti settl deal agent infrastructur html http web summari provid make exposit inform model base structur document repres contract set contract call interest element contract url refer type definit element type arrang inherit hierarchi languag interest offer abil give altern gener inherit give rang valu custom design market interest format encod interest descript framework map gener serv languag interact model asynchron messag commun speech act base market interact languag manner kqml fipa acl share common lisp base serial syntax compar similar set messag type small reject singl interest argument send messag type declin mean make legal bind requir support digit signatur authent mechan sic softwar platform agent environ softwar platform agent develop sicstu sic sicstu consist librari implement basic agent standard internet format agent os process default messag transfer messag receiv asynchron socket agent address case url type map stand market agent protocol agent messag http manner analog agent commun agent send receiv http make web client server individu web address agent default gener html intern perform automat relat approach shopbot program servic autom shop web servic jango part excit simpl interfac search bui number oper tool autom creation interfac web base simplifi task inform interact standard sic creation shopbot simpl task attent domain specif servic person servic base collabor filter techniqu fly mechan automat share base open profil standard mechan commun interest similar inform serv profil automat commun user person servic sic user inform specif provid incent inform anthoni patti mit media develop market call kasbah user assign task bui sell good perform negoti fulli automat system intend serv gener market infrastructur offer distribut gener inform interact model stanford michigan digit librari project emploi architectur system kqml influenc hierarch view system priori subdivid number compon sic attempt provid minimum glue enabl autom market particip special compon number platform build distribut commerc applic integr web ambiti eco system project aim develop architectur framework compat major internet commerc platform interact agent common busi languag insuffici inform make comparison framework present strong similar notion integr web client server object augment manner analog agent sic user screen person assist netscap commun honest car store radio fulli compat kind www figur integr user servic agent web integr user servic agent web figur illustr basic integr user servic agent web user interact servic usual web window user person assist agent augment interact interact agent servic user interact person assist dedic window section user visit servic assist detect presenc agent web page section establish connect agent servic queri assist assist initi servic specif interest enabl servic person displai offer relev current interest base notion explicitli provid user servic input respect agent format base structur document tediou creat interest form integr web enabl servic essenti part provid wai enter interest includ interact design car car creat gener interest select cd record catalog descript interest assist main goal enabl high autom internet market user deleg task handler agent handl task good offer match sell good auction user notifi event email messag cellular interest bui car settl deal car sold messag address broker directori auction pa pa figur type inform person assist agent person assist person assist inform handl autom commerc owner inform interest inform owner settl messag receiv inform store person assist figur interest interest owner market particip interest kei knowledg person assist enabl search particip match negoti price interest repres structur base templat share agent market ensur agent interest public privat public interest agent profil inform settl deal inform deal assist settl guid repres owner messag messag agent market assist agent user messag agent offer market particip messag assist inform red car sell address link agent user interact inform correspond mail address book bookmark behavior assist current rang handler handler assist handler interest number broker attempt find good offer bid process sell good act auction auction format user give interest interact run inspect result complet scenario section present coupl simpl scenario illustr person assist agent infrastructur scenario illustr person assist search bui product assum alex internet servic provid suppli person assist electron commerc alex decid bui cd mode surf servic provid start person assist contract buyer cd titl figur alex describ interest person assist alex describ interest contract buyer describ cd search pick cd templat enter inform field figur interest store public pa alex find sell cd surf directori servic pa immedi detect directori servic inform alex search directori servic gener web page summari deal interest alex time send inform directli pa dealer interest sell cd alex care pa find deal bui pa investig summari receiv directori servic start negoti seller order find deal final gener small summari offer alex find cd store offer price surf store pa inform cd interest store produc person web page full pictur music clip alex decid store reliabl examin offer final decid store choic pa settl deal pa settl deal seller addit inform address deliv cd scenario alex manual investig merchant offer told pa find bui cd automat case pa directori servic negoti automat settl deal price scenario illustr merchant autom servic sell cd assum small busi sell cd receiv larg shipment sell cd quickli contact pa statist view sale describ interest sell cd substanti lower price normal instruct pa sell pa start cd directori servic directori servic parti interest bui inform found parti cd pa wait interest parti contact cd sale figur person assist statist view time agent contact pa match interest pa investig receiv find interest match repli match interest agent initi negoti pa time price cd sold cd interest pa mark interest satisfi contact directori servic inform interest longer valid success report gener email implement develop prototyp agent system implement person assist consist core gui core agent agent system handl function implement java interfac user pa focu pa implement java netscap navig call user access pa small javascript program open navig window specif load pa gui applet applet immedi open connect pa connect open long gui run commun user pa pa gui pa gui control navig window window open javascript navig window access make load web page commerc site agent support add piec javascript code web page code open window pa gui window refer back window refer window return refer call public method pa gui applet inform servic agent navig window access content window figur show person assist browser window access directori servic code give inform broker agent locat pa broker contact pa gui var attribut function user person assist surf commerc site assist receiv address commerc agent method simpl implement problem netscap continu support kind interact navig window secur user pa gui run access agenten site window open pa gui control open window window solut problem problem solv servic provid give refer activ window inform pa gui agent web page sign applet give pa gui higher access browser web page contact pa inspect navig window check page load write client softwar window control browser altern proxi scan web page agent inform access altern attract requir softwar instal client comput discuss complement web market infrastructur greatli improv effici internet commerc market adopt solut remain problem benefit benefit larg group user creat market moder size group book market technolog ad benefit appeal group user grow larg interest target commerci incent adopt put pressur automat price comparison shop attract lead sustain growth eventu complet adopt question nich complement web fill web similar gener distribut object framework essenti softwar act behalf interact gener strong regard perform trust softwar carri routin predict review complet restrict distribut object framework acknowledg research present fund swedish transport commun research carri collabor research ab refer sic market infrastructur sic technic report swedish institut comput februari agent mediat electron trade workshop intern confer autonom inform interact usenix workshop electron commerc usenix al internet softwar platform base sicstu prolog present logic program san anthoni chavez patti mae agent marketplac bui sell good practic applic anthoni robert patti mae experi creat agent marketplac practic applic jai tenenbaum eco framework internet commerc white paper resourc descript framework yanni labrou tim finin kqml specif umbc technic report fipa acl foundat intellig physic agent agent commun languag stanford digit librari project michigan digit librari project jango shopbot excit edi kif net firefli netscap op 
IR/2521 5 gener classif evalu scheme norbert fuhr hansen michael abstract evalu digit librari essenti develop area previou approach restrict facet argu evalu dl base broad view subject area develop descript scheme major usag describ major attribut exist dl test bed perform survei mean continu set dl introduct background digit answer depend good digit research disciplin librari comput group interest make contribut digit librari view aspect relev specif viewpoint digit librari evalu identifi view approach research commun librari commun research commun inform content collect order fulfil user digit usag digit librari concern creation manag inform movement inform global network effect inform wide rang user introduct issu journal digit germani swedish institut comput sweden elsevi scienc uk hungari norwegian scienc norwai librari commun dl institut offer inform servic digit exist structur adapt technolog challeng dl provid includ offer intellectu access preserv integr ensur persist time collect digit work readili econom defin commun set commun librari feder definit dl reflect question evalu librarian focu comput scientist interest technolog aspect irrespect content dl institut subscrib dl dl similar view dl system design focu effici dl system usag comput interest effect relev appli standard inform retriev measur broader view typic task solv dl system measur task time complet rate result result manag librari select dl librarian manag subscript dl offer content system develop make design choic evalu place place system develop method perform evalu laboratori set suffic select piec softwar subscrib dl base final product state evalu meet requir set import issu construct context measur methodolog saracev conclud clear agreement element methodolog larger involv construct context evalu paper step direct resolv issu test suit group digit librari test bed develop project phase digit librari initi test bed focu project spawn test bed technolog user usag content plai minor role test bed relat work group digit librari metric form involv workshop address aspect dl evalu effort continu digit librari evalu forum delo network excel aim provid infrastructur evalu perform relat aspect access digit librari research dl larg test bed evalu demonstr concept recent excel collect creat eu collect integr applic defin describ dl test dimens test bed defin paper written member delo work group librari test give survei relat work outlin gener idea holist approach dl evalu detail present descript scheme section describ result evalu scheme ongo effort build dl give outlook futur work area relat work method tool evalu comput system investig special focu basic paramet measur perform effect effici cost factor calcul indirectli area inform retriev evalu plai central role year trec retriev initi ongo effort develop standard benchmark ir method system track trec varieti collect newspap newswir spoken broadcast interact question consid evalu retriev effect term recal regard dl point collect emploi trec lack rich structur relationship typic dl collect involv real margin interact common approach hci research area usabl evalu method process assess user interact hci set method techniqu method rang control simul situat simul task longitudin workplac studi hci point techniqu usabl inspect cognit task analysi ethnograph method order evalu usag dl usabl studi distinguish intend hard separ context evalu describ approach design dl system focu assess human inform evalu deal effect dl subsequ human inform behaviour author claim dl design iter summ digit librari similar physic differ distinguish evalu procedur physic librari digit librari review usabl evalu work taxonomi system develop addit evalu strategi evalu data collect system life includ strategi evalu method evalu evalu approach alexandria digit librari studi user reaction adl interfac function content adl user characterist studi approach current librari problem librari statist perform measur gain attent distinguish dimens cost qualiti claim tradit notion effici effect balanc benefit custom describ conceptu framework measur present list dl evalu metric user nugget serv collect qualit data relat varieti topic evalu impli qualit quantit data collect order contribut understand dl design order support user usag level system perform evalu evalu level situat contextu factor method techniqu rang quantit data collect method perform qualit data collect method ethnograph al observ call build collect evalu method techniqu framework dl evalu evalu method level analysi bibliographi holist approach dl evalu dl special kind inform consist compon comput system technic environ system built dl import integr support viewpoint dl research librarian focu comput scientist interest technolog aspect irrespect content dl ignor dimens problem lead impract solut area inform research assum evalu target interact retriev show technolog differ matter import issu consid broader viewpoint dl evalu includ underli system compon involv classic inform retriev evalu method techniqu system interfac interact level activ user system involv classic usabl evalu support access usag strategi analyt bibliograph work task support task search support design access system situat contextu factor dl group issu compar requir outcom previou dl appar imbal satisfi dl research predomin technolog relat initi order broader view dl work group decid focu develop test suit satisfi kind dl research result brainstorm develop diagram shown figur approach gener definit digit librari start illustr small circl central circl label model fall technolog definit set user predefin rang content collect arrow connect natur collect predefin rang technolog arrow attract collect user technolog user determin extent usag digit librari arrow show dot arrow show collect contribut collect technolog interact observ sociolog commun research li research user collect research domain dl domain usag technolog system research system technolog research fig schema digit librari start move outward domain dl research relationship principl research area collect creat set research requir dl test bed extent form collect predetermin rang potenti user requir technolog set evalu criteria metric categori descript scheme dl descript characterist dl interconnect depend natur user predefin collect turn delimit potenti tool set assign independ criteria domain technolog user creat definit select subset evalu criteria qualiti index thesauru citat level detail user mainten document growth object mainten interv user document inform text mine effect system structur transport model document document format user intern gener educ research distribut domain subject area distribut info seek object seek distribut consum distribut fig descript scheme evalu criteria digit librari refin concept develop previou identifi major paramet dimens dl domain paramet list figur paramet binari valu restrict set valu addit paramet numer shown ital describ paramet dimens detail user user intim connect basic question market interest behav question larg matter demograph hierarchi inform chain user intern dl system extern case extern correspond level standard inform mass market secondari tertiari educ manufactur high level corpor institut research user classif user evalu term number user type distribut user class question concern subject area interest user domain dl evalu distribut subject area metric dimens relat wai user seek question user adopt essenti strategi direct sophist tool search identifi singular piec inform resolv close defin question tradit wander approach librari systemat approach strategi user classif scheme label limit domain altern randomli wander inform light topic interest serendipit evalu distribut user approach user inform answer question user encount simpli consum inform pleasur interest inform object critic research review inform crucial work annot citat evalu distribut categori metric collect inform object digit librari content manag access qualifi collect digit librari inform object gather rule basi attribut collect themat collect work specif author js subject collect base media type ag object gener collect wide audienc collect includ varieti media type describ qualiti object manner imposs feasibl give descriptor estim qualiti authent scientif domain import collect review owner give owner collect primari object text film collect secondari object bibliograph data assist author control classif assist themat inform seek process metadata describ inform object level detail data dublin robot collect mainten redund inform object error growth collect secur bodi charg work addit function handl properli user secur access control exampl qualifi bodi mainten statist growth number type technolog technolog issu subdivid user inform system structur document technolog user technolog deal function dl system offer function provid user interfac document made access search disclosur mechan notifi user document relev relev document user prefer read print function essenti user work support user group import function dl collabor filter access exist dl system support creation document inform dl system implement rich set function retriev search document queri navig link document metadata base profil filter locat potenti relev document stream incom document inform extract gener fact text document base text mine discov correl trend document collect system structur technolog deal architectur repositori manag transport model commun system user interfac system document technolog address issu document document model describ abstract structur document logic content semant content extern attribut document format specifi syntax intern document evalu questionnair lesson learn scheme serv basi evalu classif effort attach answer question descript digit librari receiv descript compar digit librari select specif test trial survei half questionnair reflect main idea classif scheme survei order gather set inform digit librari collect test design questionnair questionnair digit librari test concern avail exist digit librari test collect research field digit librari questionnair digit librari test investig futur requir digit librari test collect question questionnair put similar word chang reflect exist dl versu requir research set question correspond main categori technolog gender work domain person respond answer question select singl multipl choic choic list option give comment respond answer question questionnair question questionnair question reason questionnair question question relev put exist digit survei carri survei tool survei announc mail list major audienc digit librari research estim roughli target audienc respond respond research user evalu digit librari doubl weight research domain domain survei show classif scheme dl word question problemat due holist approach classif research area area term set languag make difficult creat questionnair equal understand research area creat question live multidisciplinari research area easi task receiv suggest comment survei show addit problem guess reason granular choic hand hide great import research survei low rate show return invest research commun possibl fill survei start point research cooper survei environ requir level dl current effort delo work lesson learn survei idea account extens survei dl regist provid inform inform provid updat time origin authent question survei redesign opportun answer free text question map hierarchi suggest classif scheme node hierarchi repres research area dl function call slot identifi uniqu number sequenc question answer store question easili introduc exist survei slot enhanc search content user hierarchi summari statist solut select search dl solut select slot suffici coverag activ dl peopl find research partner time lack featur hole research identifi outlook argu evalu dl base broad view subject area develop descript scheme major dimens major properti dimens exist dl test bed perform survei mean continu set dl order mainten investig possibl switch harvest provid dl test bed maintain metadata record describ test harvest collect record regular basi metadata substanti number exist test questionnair desir test exist test bed current interest dl type test bed outcom poll lead effort build test similar trec initi acknowledg work support part network excel delo fund european commiss refer global inform access inform network world mit user evalu alexandria digit librari project inst imposs thing approach measur librari proc intl conf perform measur librari inform page newcastl inform north servic qualiti academ librari ron jame rb dr anna simpson user summari methodolog result alexandria digit librari proc page inform todai public servic survei poll proc dexa page septemb lamb onlin inform reach digit librari confer acm landauer troubl product mit peopl digit multifacet approach assess impact digit librari social practic design evalu mit nielsen rl editor usabl inspect method john wilei catherin plaisant anita evalu challeng feder heterogen inform case earth scienc inform partnership proc wet ic ieee ws evalu collabor carei comput interact addison saracev challeng digit librari evalu volum page harman overview eighth text retriev confer eighth text retriev confer page 
ML/1510 6 brock depart comput scienc genet program definit translat grammar brian ross tech report june brock dept comput scienc st ontario canada genet program definit translat grammar brian ross brock dept comput scienc st canada ext abstract genet program system definit translat grammar logic version attribut grammar support definit semant inform languag easili accomod grammar genet program defin interpret target incorpor syntact semant contraint evolutionari search system improv gp system permit semant aspect languag defin grammar automat analyz grammar rule order determin minim depth termin requir gener random program tree vari shape size applic introduct genet program implement benefit simpl program denot koza denot program tree implement simpl languag nontermin basic grammat definit program structur make program gener reproduct implement effici execut gp system grammat requir express singl singl nontermin argument defin potenti advantag scheme permit rich search space wide syntact varieti tree permiss implementor ensur closur program execut error simpl treatment program structur detriment gp system requir branch program tree special case complic implement adf definit requir special denot adf branch research suggest rigor program tree data type grammar benefici evolut effici reason basic necessarili favour semant program evolut effect critic target languag restrict syntax lisp grammar permit problem domain specif syntact constraint inject evolv program boost evolut effect refin search space system genet program system definit translat grammar logic grammar permit style specif grammar permit multipl semant definit includ definit syntact definit languag unifi semant properti benefici gp system number wai complet comput semant languag defin order defin complet environ separ interpret compil semant rule defin semant properti simplifi target languag grammar syntact constraint aid semant refin constraint conveni depend partial interpret program order determin tree gener reproduct process evolut gp system gener random tree initi popul creation program mutat size tree control random tree ensur tree varieti shape initi popul control tree size gener gener grammar overcom grammar analyz grammar rule characterist rule minim depth gener rule inform requir effect gener program tree desir depth initi popul program mutat section review genet program definit translat grammar system present section applic section compar gp system section section conclud paper discuss background genet program genet program method automat program program evolv genet algorithm gp ga gener initi popul loop current gener maximum gener fit individu consid loop popul size maximum popul size ffl select genet oper select individu base fit perform crossov add offspr popul select individu base fit perform mutat add offspr popul ffl increment gener counter print solut obtain figur genet algorithm figur initi popul random case gp finit number result replenish popul depend fit individu numer score indic individu abil solv problem select program select reproduct proport reproduct crossov mutat select gener offspr gener essenti differ gp ga denot individu popul pure ga genotyp decod phenotyp problem solv gp comput program directli execut interpret program code genotyp gener permit vast number problem solvabl mean gp closest realiz true automat comput program descript problem solv main reproduct oper gp crossov mutat crossov import permit genet combin program code program act mean inherit desir trait evolut crossov select find random crossov point program intern swap crossov point crossov point select result offspr syntact correct koza simpl program structur mean nontermin root select swap nontermin root program mutat commonli perform gp random mutat point found select node replac randomli gener tree mean genet trait introduc popul evolut crossov mutat preserv grammat integr user ensur closur result program execut long closur program deriv gp system execut fit fit deriv definit translat grammar invent mean syntax semant languag unifi fashion syntact possibl multipl semant compon syntact compon definit grammar permit definit languag style syntax logic grammar implement prolog syntact compon permit prolog call semant rule nontermin product semant refer repres inform verifi grammat construct semant viabil gener definit inform fundament formal grammar product form start symbol deriv step repres applic product fl nontermin deriv conveni denot deriv tree root intern node leaf descend node repres applic product nontermin node note nontermin multipl product applic logic program goal nontermin termin refer side product set goal product bodi product includ refer nontermin product option data argument variabl denot place holder data structur element head goal syntact form rule line head defin syntact definit nontermin encod bs goal product refer termin prolog list call prolog refer semant compon case goal brace symbol delimit syntact rule semant definit head semant definit product goal refer semant goal prolog code goal semant specif refer semant goal syntact goal syntact product variabl link semant goal syntact goal goal semant definit refer semant compon tree nontermin syntact definit product semant scheme permit access semant compon typic product natur languag process rule verb phrase syntact defin singl cfg product state verb phrase verb noun phrase interven refer transit prolog verifi verb refer transit rule interpret user evalu true sentenc consid grammat correct semant rule product rule gener agreement class verb base agreement class verb singular plural logic form rule gener logic portion detail call logic form verb noun phrase portion tree system design genet program system implement prolog silicon graphic irix window platform librari author harvei engin convent gp gener tournament local search paramet gp control outlin section remaind section discuss featur grammat logic foundat grammar syntact semant definit grammar initi librari util translat standard prolog argument list token rule grammar permit gener sentenc conform grammar intern tree structur repres tree program nontermin node node leaf node repres nontermin structur repres singl intern node tree call copi head grammar product exact rule identifi id number list node descend node encod product node node includ refer semant goal prolog code goal goal rule interpret parallel interpret tree call parallel interpret goal fail tree tree gener reproduct step fail main gp engin evolutionari action undertaken grammar preprocessor gener random select rule deriv gener random program shape size semant goal prolog goal rule bodi repeat rule process progress ffl examin rule rs bodi made termin min rule nontermin rs bodi minima min rule maximum minima nontermin label min found time step min min depth nontermin rule bodi test step succe put back iter figur minim depth algorithm execut random act addit verif tree construct tree gener initi popul mutat step evolut gp run commenc gener initi popul random program convent wisdom evolut effect initi popul program varieti shape size popular mean deriv popul koza ramp gener strategi half program gener tree gener rang maxim branch tree maxim depth half popul rang tree depth tree branch greater maxim depth gener tree desir depth easi koza grammar grammar specifi express nontermin express argument deriv termin termin nontermin deriv tree desir depth trivial grammar tree nontermin node expand product nontermin difficult ensur tree desir depth directli determin applic product result branch requir depth product termin immedi bodi consist termin product infinit problem gener grammar number investig problem solut share characterist simpl algorithm yield accept albeit complex approach gener ramp popul analysi grammar order determin characterist minim depth tree gener rule gener termin arbitrarili deep tree characterist rule select prudent characterist identifi rule termin branch characterist identifi rule result tree fuller branch analysi determin minim depth gener grammar rule multipl rule applic rule gener vari minim tree depth gener tree select rule assur gener tree depth limit algorithm comput minim rule depth figur iter algorithm process grammar rule set fashion time minim depth determin minim depth nontermin refer rule base case rule leaf minim depth iter case rule nontermin minim minim depth rule level deeper maximum minim depth process continu rule minimum depth comput rule undetermin minim depth discov rule mean fix user analysi determin termin characterist rule rule descend result nontermin deriv finit tree rule knowledg properti rule gener full requir branch deep order gener full random select rule enabl deepest branch select rule gener finit depth tree forc termin shallow depth desir rule descend rule gener tree requir maxim depth algorithm determin termin properti rule figur algorithm case attempt label rule label continu label rest unlabel label user overrid termin design gener deep program structur languag construct gener deep pascal set structur gener set includ member space method determin minim rule depth rule base sole nontermin label analysi product algorithm give repeat rule process progress ffl examin rule bodi made termin nontermin bodi label termin label termin nontermin label nontermin label nontermin nontermin rule label nontermin test step succe put back iter ffl label remain unlabel rule nontermin figur rule termin algorithm approxim minim size tree deriv natur product actual minim depth product product nontermin applic due exist argument semant prolog goal rule featur ignor sake effici practic design support fact argument valu depend time data popul gener main goal tree gener desir shape reproduct oper major crossov mutat oper offspr grammat correct respect origin suggest maintain grammat integr crossov ensur crossov select result grammat correct offspr select parent root nontermin type termin select unlabel easili identifi correct user select termin inform nontermin rule defin nontermin root select mutat tree nontermin type select replac select product semant goal prolog interpret crossov mutat order verifi correct tree consid verif step perform offspr applic syntact semant rule prolog semant verif unnecessari omit crossov mutat fail number wai nontermin type select parent exist parent offspr exce maximum depth bound semant verif fail situat reproduct step retri parent number time reproduct oper fail select evolut continu select parent reproduct experi section present applic evolut probabilist pattern match languag experi spirit work consult detail overview problem mean probabilist languag implement discuss illustr main benefit paramet evolut strategi result experi assum reader background formal languag probabilist pattern match grammat infer classic problem machin learn ga gp successfulli appli infer formal languag ga gp stochast languag infer investig stochast languag formal languag probabl distribut set member point addit burden infer languag membership probabl distribut member ascertain probabl distribut advantag infer algorithm number reason probabl preclud neg probabl posit member automat account neg membership probabl distribut lend addit freedom lend natur mean statist accuraci accept solut probabilist languag infer natur suit gp effect problem requir accur good correct latitud error target languag probabilist regular call stochast regular express similar stochast regular languag convent regular express regular oper numer field indic probabl produc result languag number notat express stochast regular stochast finit automata call hidden markov stochast regular grammar notat theoret convert stochast regular natur express notat simpl express describ complex stochast regular vice versa linguist express denot impact effect infer complex solut natur requir effort infer algorithm syntax ff rang alphabet rang rang posit integ rang decim valu decim place syntax defin ff loss empti string ffl includ alphabet languag express mean oper atom action action ff gener probabl guard choic ff ff ff term choic express prefix uniqu atom action found express make guard choic contribut smaller express gp process compar nondeterminist choic term probabl express term probabl probabl probabl entir express product probabl chosen term probabl express encapsul concaten term probabl product probabl express kleen closur term repeatedli execut time iter occur probabl multipli probabl probabl iter express denot set string belong probabl distribut languag interpret probabl field inform explain oper descript express string member probabl term chosen probabl term iter probabl final iter termin probabl string member probabl string ba member probabl definit syntax expr expr expr expr expr iter choic choic choic expr figur syntact definit figur encod grammar elucid simplifi remov semant rule rule variabl semant rule rule function rule expr rule defin guard express order includ kleen closur rule refer semant goal construct nontermin choic construct list prefix term choic express call prolog goal uniqu guard succe prefix uniqu tree fail due instanc duplic prefix rule encod addit syntact constraint directli nest iter permit prior experi found nest iter commonli construct gp unfortun effect dramat slow interpret time gener minut probabl due compound applic iter probabl disallow significantli speed neg effect evolut rule simpl check semant express express rule fail rule defin defin guard express term notat succe rule termin member alphabet rule defin integ float point choic probabl iter abil grammat constraint permit implement worth examin contribut semant rule detail rule semant check nest iter inform express nest rule iter determin tree semant complex grammar requir expr expr iter expr iter involv simpl check rule altern unwieldi separ set rule requir order captur action process guard express attribut grammar process guard choic figur choic choic choic choic choic choic strategi grammar small alphabet definit oper semant choic figur oper semant choic oper section evid fairli oper semant oper oper oper semant conveni encod semant field fit gp test express recogn string target languag oper semant encod express interpret attempt recogn membership produc probabl recogn string actual implement find probabl largest prefix recogn express complet account implement oper semant scope paper give flavour figur semant rule figur rule pertin choic occur end tree choic choic rule string start process choic string process equal suffix sum probabl valu term choic list denomin probabl comput process current string express compon read earlier prefix final comput probabl pr process complet rule rule process term pair choic express call construct retriev actual integ probabl field term probabl term multipli probabl intermedi probabl pass recognit semant express choic term rule invok multipl wai string recogn express oper semant exhaust rule string complet altern deriv express probabl obtain differ deriv path collect yield probabl string greatest prefix case experi detail tabl paramet paramet target languag termin root expr fit function modifi random test set gener type initi popul size run popul size uniqu popul member maximum gener probabl crossov probabl mutat probabl intern crossov probabl termin mutat probabl numer mutat numer mutat rang max reproduct attempt initi popul shape ramp depth initi max depth offspr tournament select tournament replac test set size approx max test string size probabl limit task evolv express correspond languag target fit strategi test express recogn random test set string gener target descript test set randomli gener entri result test set list string valu frequenc test set string interpret encod membership probabl comput modifi formula fit evalu jt jt frequenc test set jt maximum prefix recogn term string complet recogn formula prefix recogn proport size prefix recogn normal formula prefix score credit express recogn portion drive evolut express recogn complet exampl gp paramet list figur gp literatur mutat special mutat numer termin found express probabl numer mutat state time numer field select mutat rang specifi numer randomli perturb fail reproduct oper nontermin match select parent probabl limit specifi lower cutoff probabl recognit probabl lower recognit current express deriv path deriv attempt boost effici deriv result tabl summar experi result valu tabl comput run fix test set note variat fit individu test randomli gener test set vari gener run averag test set indic test set vari averag solut obtain set run solut good figur show averag perform run tabl summari total run uniqu exampl avg test set fit min max avg fit translat readabl probabl choic directli simplifi remov term statist essenti target languag express note remov express term materi genet inform ill effect fit size poor solut fit transform manner term accur process prefix target languag partial termin occur correctli probabl work inspir system implement logic program logic grammar paradigm crossov procedur blindli select node potenti gener fit averag figur fit curv succe determin offspr grammat correct crossov proport crossov oper fail overcom select similarli label node type crossov detail system program logic grammar argument structur maintain crossov reproduct oper simplifi tree structur denot essenc grammar tree semant compon requir verif crossov grammar interpret grammar encod program permit semant execut circumv requir major advantag addit semant express inher motiv place encod substanti semant inform grammar overli complex unwieldi overcom weak encod semant separ ty rule pertain argument notat permit semant inform goal syntact rule extract requir net result complex includ entir languag encod semant definit practic incorpor syntact constraint grammar target languag syntact constraint defin syntact semant greatli simplifi paper gruau cfg enforc syntact properti genet program user syntact constraint solut encod properti grammar target languag search space semant search space constrain user semant constraint effect control syntact constraint gener mean control search logic grammar gener constraint gruau system call count ensur program tree exce depth limit ensur rule select program gener result unduli small identifi termin properti rule rule gener program initi popul tend shallow lack control gener program size crossov system select nontermin grammat correct gruau transform grammar normal permiss crossov possibl product gruau binari encod grammat resourc argu encod hinder execut speed cheap memori cfg gp gruau system node nontermin identifi select crossov exchang control size gener tree requir user manual encod product number indic minim tree depth rule automat determin inform analyz termin characterist initi popul favour smaller tree research suggest linear gene number map grammar rule gene map perform legal grammat tree index map erron grammat structur translat denot arguabl akin ga chromosom convent gp translat step basic genotyp phenotyp undertaken grammat correct preserv erron chromosom map circumv gene translat grammar rule tree gener list depth tree termin characterist grammar rule consid implement ga gp system grammar formal grammat gp jacob grammat gp encod grammar natur graphic model natur phenomena plant structur attribut grammar gp evolv neural network work relat grammat gp data type gener grammat data type constrain solut respect rang argument valu pass program construct motiv principl pure grammat constraint problem gener random tree size shape constraint investig number algorithm algorithm basi work linear algorithm gener random grow tree gp show basic grow tree gener algorithm produc uniform distribut tree rang depth tree gener algorithm gener tree uniform distribut shown posit influenc evolut algorithm design koza basic cfg directli applic complex cfg bohm address random tree gener grammat gp exact shape depth metric tree size number deriv step requir grammar deriv tree combinatori view languag tree gener set number deriv step belong partit languag search space algorithm knowledg word partit randomli deriv tree partit deriv tree varieti tree randomli gener admit implement algorithm partit space depth breadth characterist tree algorithm paper simpl result accept variabl random tree shape paper discuss implement gp system encod syntax semant target languag build earlier extend system borrow idea maintain grammar correct reproduct system analyz user grammar inform termin characterist requir effect random tree gener illustr practic semant simplifi grammat definit encod oper semant target languag languag composit oper defin interpret procedur forward syntact problem constraint defin grammat definit unwieldi result tree program syntact properti grammar semant substanti simplifi grammar semant encod complex semant properti interpret partial program order determin accept benefit prune search space manner outweigh result overhead perform reproduct search space evolut success investig semant constraint gp number enhanc gruau normal form product worth normal form transform ensur termin nontermin ancestor transform perform robust opportun node select direct crossov experi evolv product evolut suggest grammar evolv order discov instanc work attent direct substanti bodi work grammar prolog lisp ai symbol paradigm make ideal genet program implement slow compar gp system program compil languag interest interfac gp system implement faster languag work support nserc oper grant refer dahl logic grammar alonso random gener tree kluwer academ angluin comput learn survei select bibliographi annual acm symposium theori page acm fd genet introduct morgan bohm exact uniform initi genet program rk belew foundat genet algorithm morgan brave evolv determinist finit automata cellular encod john koza proc genet program page stanford morgan kaufmann charniak statist languag learn mit wf cs mellish program prolog regular grammat infer posit neg sampl genet method intl coll grammat infer page jj freeman linear gp context free grammar jr koza proc genet program page morgan ks fu tl booth grammat introduct survei part ieee transact januari vk si marcu probabilist languag framework stochast discret event system technic report institut system april fuzzi expert system genet machin learn studi fuzzi soft edit program languag genet program john koza proc genet program page stanford morgan kaufmann gruau syntact contraint genet program pj angelin ke advanc genet program page mit td da rl type inherit strongli type genet program pj angelin ke advanc genet program page mit jh holland adapt natur artifici system mit je hopcroft jd ullman introduct automata comput addison ts ra attribut grammar genet neural network syntact constraint genet program workshop evolutionari iba random tree gener genet program parallel problem solv natur proc intern confer evolutionari page jacob evolv evolut genet program jr koza proc genet program page mit te rk belew stochast grammar induct genet algorithm local search rk belew foundat genet algorithm iv jr koza genet program mit jr koza genet program ii mit mm grammat infer genet algorithm confer massiv parallel process applic page evolutionari learn larg grammar jr koza proc genet program page stanford morgan kaufmann luca structur chromosom grammar evolut intern confer evolutionari page ieee dj montana strongli type genet program evolutionari pereira warren definit grammar languag analysi survei formal comparison augment transit network artifici lr rabin tutori hidden markov model select applic speech recognit februari bj ross probabilist pattern match evolut stochast regular express technic report brock dept comput june jj oneil grammat evolv program arbitrari languag banzhaf proc european workshop genet program page sakakibara recent advanc grammat infer theoret comput infer stochast regular grammar massiv parallel genet algorithm proc intl conf genet algorithm sic sicstu prolog user june learn regular languag genet program jr koza proc genet program page morgan pa genet program jp workshop genet theori page pa induct bia genet program intern confer genet algorithm engin innov applic page ml wong ks leung evolutionari program induct direct logic grammar evolutionari zhou jj grefenstett induct finit automata genet algorithm proc ieee intl confer page ieee press 
ML/2573 6 kluwer academ boston manufactur netherland cognit bia approach featur select weight learner clair cardi depart comput cornel ny david aha abstract research cognit scienc discov examin numer psycholog constraint human inform process short term memori focu attent prefer tempor recent inform exampl paper show psycholog constraint sourc bia guid featur set select weight learn algorithm show cognit automat explicitli encod baselin instanc bia modifi chang delet modifi featur weight investig relat problem cognit bia select cognit bia interact featur weight approach compar algorithm bia select make assumpt independ individu compon evalu natur languag learn show bia select algorithm determin cognit bia relev learn task accuraci learn algorithm improv significantli select incorpor baselin instanc featur set featur natur languag learn introduct induct concept acquisit primari interest research field machin learn cognit scientist examin numer psycholog limit human inform process fact concept learn basic cognit cognit process limit rare exploit design machin learn system concept acquisit paper show cognit process limit sourc bia guid featur set select improv learn algorithm perform describ cognit automat explicitli encod train instanc simpl learn algorithm neighbor initi focu singl learn task field natur languag process present baselin instanc modifi cognit focu attent bia recenc bia short term memori limit seri clair cardi compar modifi instanc baselin descript find cognit bia significantli improv system perform addit gain accuraci achiev appli cognit simultan baselin instanc psycholog process limit includ instanc system address relat issu cognit bia interact cognit bia select paper present method cognit bia select make vari assumpt independ individu process limit method combin search cross valid greedi approach bia select increment incorpor baselin individu cognit learn algorithm accuraci continu improv method assum bia interact algorithm bia select make assumpt bia interact exhaust evalu combin cognit work previou work search cross valid bia select featur select direct featur set select featur weight result experi show bia select algorithm determin cognit relev learn task perform learn algorithm improv significantli select bia incorpor baselin instanc investig gener cognit bia approach featur set select show addit cognit translat baselin instanc appli featur select algorithm cognit addit natur languag learn task find cognit bia select algorithm incorpor relev significantli improv learn algorithm perform remaind paper organ section describ natur languag learn task rel pronoun disambigu baselin instanc task paper introduc compon cognit bia approach featur set select section present learn algorithm evalu rel pronoun task method incorpor independ primari cognit focu short term memori limit section section evalu altern approach bia select briefli outlin evalu cognit bia approach featur set select addit data set section cardi describ implic work natur languag process machin learn cognit bia approach featur select rel pronoun disambigu goal machin learn algorithm natur languag task disambigu sentenc toni boi won award learn algorithm locat phrase repres descript context occur sampl system recogn refer find rel pronoun crucial task natur languag understand system part fill semant role sampl object implicit actor focu disambigu rel pronoun occur frequent sentenc disambigu rel pronoun determin critic larger inform extract task learn algorithm exist natur languag process system includ disambigu heurist directli compar learn algorithm larg bodi literatur human process rel corpu broader inform extract task section focu specif learn disambigu heurist frequent rel pronoun occur ten sentenc higher frequenc import text relev inform extract task major psycholinguist studi human process rel focu find rel pronoun simpl factor make head noun rel pronoun consist syntact constitu posit exampl figur direct object subject head phrase immedi case show case head distant rel pronoun conjoin noun phrase conjunct phrase semant valid refer entiti clair cardi entir phrase similar situat occur predic nomin construct appar sentenc analyz distinguish instanc true rel pronoun locat requir assimil syntact semant knowledg syntact structur sentenc ident case head noun preposit show machin learn system learn locat descript toni boi won award boi gave book red hair toni dinner men detroit sold comput spoke woman black shirt green hat corner room interview id provid id provid support watch woman philadelphia plai soccer sister gift children attend parti tabl figur baselin instanc circu sentenc analyz gener train instanc system gener instanc occurr text muc terror corpu corpu develop conjunct messag understand confer perform evalu inform extract system inform extract system input unrestrict text text respect prespecifi topic domain find inform domain encod inform templat suitabl popul circu system consist strong perform muc evalu muc collect consist document includ newswir radio tv text text upper case cognit bia approach featur select train instanc rel pronoun task list pair encod context found train instanc annot class describ posit correct class featur predict learn algorithm test detail baselin instanc kei characterist sentenc circu parser recogn phrase find sentenc recogn major constitu direct object make structur attach handl preposit phrase attach circu semant featur describ noun adject lexicon noun attack circu track recent recogn punctuat mark rel pronoun disambigu case type phrase case pair describ phrase case tabl phrase constitu featur denot syntact class posit phrase encount parser featur parser recogn subject exist subject declar preposit phrase subject sem featur phrase semant classif system lexicon sem featur subject specifi locat phrase preposit includ prep featur denot preposit featur phrase punctuat mark conjunct mark featur includ denot mark featur comma phrase featur instanc denot syntact type linguist entiti immedi sampl featur comma comma conjunct phrase construct label separ sentenc tabl direct object conjunct clair cardi tabl rel pronoun resolut baselin instanc man phrase featur man tabl rel pronoun resolut baselin instanc phrase featur instanc recogn direct object noun phrase direct object verb phrase semant featur sem featur verb exampl tabl illustr import characterist rel pronoun data featur case structur vari rel pronoun occurr train instanc annot class inform list constitu attribut repres posit present sentenc tabl phrase repres class conjunct constitu repres list constitu attribut element conjunct sentenc tabl refer conjunct encod apposit predic nomin construct result set semant part class inform sentenc combin phrase cognit bia approach featur select consid correct option predict evalu baselin instanc section describ learn algorithm evalu baselin case rel pronoun disambigu learn algorithm emploi simpl learn algorithm albert train rel pronoun instanc simpli store case base rel pronoun weight neighbor case retriev algorithm predict compar test case base calcul wn yn denot test case wn weight ith featur xn featur test yn train function return equal baselin wn return case highest score singl case class inform find test case retriev case vote posit case direct object test case sentenc select retriev case list option apposit predic nomin option list constitu overlap current case retriev algorithm match featur test case retriev algorithm normal featur set train case match respect expand featur set obtain compar perform rel pronoun task normal featur discuss result clair cardi rel pronoun data set rel pronoun data set instanc circu gener case occurr text corpu correct case human supervisor access version train corpu annot rel attach inform experi made pass data set correct small number obviou semant class disambigu error perform learn algorithm underli characterist data set instanc featur repres phrase case uniqu class distinct valu data set ten instanc uniqu valu establish upper limit accuraci baselin instanc data set pair instanc set instanc context lack lexic featur distinguish type pronoun case involv syntact ambigu construct respect rel pronoun resolut includ sentenc head noun constitu constitu syntact viabl walk learn algorithm semant class inform punctuat correctli handl case small number case remain ambigu addit context result result learn algorithm baselin case shown tabl weight set mention algorithm perform calcul test case featur normal featur set result paper cross valid averag statement statist confid level mcnemar test statist row tabl show accuraci baselin learn algorithm remain row tabl show perform addit baselin system system implement default rule rel pronoun disambigu heurist recent constitu resort complet constitu expect system specif human cognit bia approach featur select perform result recent human default rule show case problem legitim semant featur human complic semant featur combin rule recent human proper simplest default rule tabl result baselin default featur weight algorithm rel pronoun resolut baselin instanc recent constitu recent human recent human proper heurist fourth baselin system tabl emploi set heurist rel pronoun resolut develop perform evalu heurist consist approxim rule verb subject constitu head constitu verb token head subject rule make syntact semant inform encod baselin instanc origin base approxim instanc rel pronoun muc terror modifi period handl encount test full inform extract system result tabl baselin system perform default rule heurist mcnemar test system indistinguish term statist accuraci baselin recent constitu default rule behavior qualit baselin perform markedli default rule immedi rel perform markedli default rule complex subject detect rel pronoun final row tabl explain section clair cardi comparison section compar perform baselin learn algorithm featur weight algorithm shown good perform number natur languag learn task weight neighbor algorithm straightforward composit exist approach featur select featur weight tree featur select cardi briefli goal step prune featur learn algorithm ignor assign remain featur weight inform gain train case algorithm van den intent weight featur rel import data set step train creat case base simpli store train instanc train instanc creat tree learn task experi comput featur weight case retriev comput tree step inform gain ratio comput train instanc appli algorithm rel pronoun data normal instanc respect entir featur fill nil miss featur class instanc determin weight case retriev algorithm ident baselin learn algorithm featur weight comput tabl approach work poorli rel pronoun task work natur languag learn problem includ semant class preposit phrase noun phrase chunk daeleman work poorli larg number class inform gain bia rel pronoun normal instanc miss valu perform data set section incorpor cognit modifi baselin cognit measur learn algorithm abil predict rel pronoun cognit bia approach featur select incorpor subject access bia number studi psycholinguist note special import item mention sentenc shown access actor sentenc remain high end sentenc subject access bia process rel note king gener focu attent bia comput vision learn object view highli access object learn aural loud sound highli access incorpor subject access bia baselin case weight featur subject rel pronoun weight subject featur function fix subject weight subject weight divid evenli featur subject possibl ad origin weight subject featur tabl subject access bia rel pronoun resolut baselin subject wt subject wt subject wt subject wt tabl show rel pronoun resolut subject weight incorpor subject access bia improv rel pronoun disambigu compar baselin dip perform statist test mcnemar test baselin weight higher subject weight provid improv perform result baselin produc circu encod subject access bia explicitli recogn subject major constitu sentenc label noun phrase bia baselin drop perform origin encod bia adequ addit modif baselin requir subject access bia posit subject access bia small number featur case subject featur subject identifi rel pronoun subject access bia plai role analysi error baselin perform slightli major type clair cardi tabl incorpor recenc bia label baselin sentenc label congress incorpor recenc bia process peopl consist show bia recent inform kimbal nicol gibson gibson al fodor mitchel investig import recenc find rel pronoun found english prefer recent noun phrase sentenc rel pronoun ambigu sentenc journalist interview daughter peopl assum refer daughter featur select algorithm translat recenc bia train test instanc wai direct modif featur modifi weight constitu distanc rel pronoun renam featur posit constitu rel part subject relabel noun phrase verb unambigu denot presenc subject establish label constitu label natur parser baselin instanc tabl show label leftmost rightmost column origin baselin recenc label attribut preposit phrase posit left origin phrase label respect direct object receiv attribut recenc noun phrase posit order yield featur set instanc uniform encod context rel pronoun cognit bia approach featur select tabl incorpor recenc weight maximum weight phrase attribut baselin weight recenc weight final weight congress messag congress congress point view rel pronoun sentenc similar refer case label assign recent constitu sentenc attribut assign class instanc constitu baselin label recent constitu instanc distinct make case retriev approach incorpor recenc increment weight featur function proxim creat weight featur element closest rel pronoun receiv weight ten weight featur reach recenc weight ad origin baselin featur weight produc final weight vector nlp system gener make implement recenc weight assign weight featur singl phrase altern work recenc weight scheme nlp system rel pronoun data set result experi recenc separ combin form shown tabl combin implement recenc system relabel attribut instanc label initi weight vector recenc weight procedur tabl show recenc weight label predict appli perform run baselin statist combin improv perform significantli respect baselin instanc recenc weight clair cardi tabl recenc bia rel pronoun resolut boldfac entri respect baselin recenc weight baselin recenc heurist label weight mcnemar test respect standalon label combin recenc bia perform recenc effect strong rel pronoun resolut individu implement recenc bia complement local context provid label critic find recenc weight lack local addit emphasi constitu closest rel pronoun broad instanc space recenc label origin data set characterist appli bia section data set encod baselin case exhibit distinct incorpor recenc number reduc number instanc uniqu similarli reduc ten spite reduct data set number instanc type remain uniqu baselin analysi test combin recenc bia correct baselin true ten time combin markedli baselin recogn rel pronoun combin perform baselin distanc rel half involv distanc phrase rel pronoun instanc combin recenc correct baselin list denot phrase select subscript mean baselin combin recenc select phrase compon awar rid baselin learn kill combin recenc jose maria martinez roberto combin recenc italian rossi combin recenc stai cognit bia approach featur select tabl restrict memori bia rel pronoun resolut baselin memori limit memori limit memori limit memori limit incorpor restrict memori bia psycholog studi determin peopl rememb minu item time carpent show work memori capac subject abil find refer pronoun vari distanc king show work memori capac read time class rel languag learn human success limit inform process capac children ignor linguist data receiv comput languag learn system elman build short term memori directli architectur system clear baselin instanc rel pronoun task make short term memori learn algorithm featur case retriev short term memori explicitli state short term memori limit vari depend cognit task depend size type rememb short term memori bia state featur king exist interact short term memori limit attent subject access semant sentenc remain activ long distanc argu implement restrict memori bia accommod influenc cognit memori limit restrict memori bia select featur highest randomli case ti set weight remain featur prune select featur instanc baselin featur weight restrict memori bia discard relev featur discard irrelev expect bia posit impact perform combin cognit provid addit featur relev inform version restrict memori bia select featur phrase select featur altern implement presenc cognit modifi weight individu featur entir constitu clair cardi tabl show result restrict memori bia set random compon restrict memori result averag clear tabl bia degrad abil system predict rel pronoun drop perform statist test test show perform respect result random featur select restrict memori bia appli isol restrict memori bia depend number featur test case memori limit averag number featur instanc maximum number featur instanc number test case featur memori restrict memori bia limit perform approach baselin discuss result tabl individu cognit bia summari rel pronoun resolut boldfac entri statist respect baselin heurist cognit bia baselin system paramet correct baselin system default rule heurist baselin singl subject access subject restrict memori memori recenc recenc max combin bia recenc recenc max implement tabl summari variat bia implement baselin system cognit bia implement significantli improv accuraci rel pronoun predict baselin subject access memori boost perform perform recenc weight label appli isol statist combin implement recenc bia recenc weight label featur obtain produc perform gain respect baselin case combin recenc significantli outperform heurist section present evalu automat method select combin cognit bia approach featur select divid data partit partit test test test partit train test train data divid train partit partit train train select data train train train learn data appli cognit bia combin paramet train turn test train ii bia combin comput averag accuraci select data partit iii select bia combin highest averag accuraci appli select train test test return averag accuraci test partit figur algorithm bia select greedi exhaust bia select algorithm method instanti boldfac step tive expect perform complimentari cognit bia select section show spite result human languag cognit appli isol found gain perform rel pronoun resolut task achiev appli cognit bia instanc implement recenc bia instanti simultan experi cognit bia interact make determin relev learn task combin recenc perform spite small gain produc recenc bia isol present autom method cognit bia select greedi bia select exhaust bia select make assumpt independ individu cognit approach reflect potenti qualiti select bia combin comput time requir make select bia select algorithm high greedi exhaust bia select method emploi nest cross valid suggest outlin figur clair cardi greedi select accuraci find remain bia appli accuraci exit incorpor drop accuraci return incorpor remov variat return figur greedi bia select algorithm instanti boldfac step figur bia select algorithm li method instanti highlight step gener algorithm step determin structur bia space order search greedi bia select algorithm oper incorpor remain individu accuraci select data improv remain constant rel pronoun data paramet includ cognit varieti paramet combin recenc max wt subject access wt iter select incorpor instanc altern variat newli select bia remov set cognit bia approach featur select maximum weight ten bia recenc delet iter bia select exhaust bia select algorithm oper simpli test combin paramet set exhaust approach bia select instanti boldfac step gener bia select algorithm initi set cognit paramet creat combin thereof bia combin appli bia combin learn data test select data comparison greedi bia exhaust bia select make assumpt cognit bia method requir comput time select bia combin data set test cognit paramet set greedi bia select test set combin exhaust requir test combin merg bia greedi exhaust bia select algorithm requir order procedur merg produc individu greedi accomplish incorpor bia relabel add attribut incorpor modifi featur weight ad weight vector bia recenc subject incorpor discard featur restrict memori case includ cognit combin bia creat automat user specifi list appli problem paramet result tabl show result appli bia select algorithm rel pronoun disambigu task level cross valid ten partit tabl list select number partit combin select consid shown bottom tabl greedi algorithm make fairli uniform select cognit partit combin recenc bia select restrict memori bia clair cardi tabl bia select result rel pronoun resolut task bia select tabl show select number cross valid partit bia select accuraci refer averag correct ten partit select number partit greedi select rec combo wt rm rec combo wt rm subj rec combo wt rm rec combo wt rm subj rec combo wt rm subj exhaust select rec combo rm subj rec combo rm subj rec combo rm subj recenc label recenc weight max wt combin max wt subject access wt restrict memori limit select memori limit rang subject access bia select ten partit subject weight case averag accuraci creat greedi bia select significantli outperform baselin case default individu cognit significantli outperform combin recenc exhaust bia select fairli stabl set cognit select combin recenc maximum weight restrict memori bia small memori limit subject access bia small subject weight averag accuraci creat exhaust bia select greedi bia exhaust bia select perform significantli baselin case default individu combin recenc test perform compar greedi select algorithm run time partit exhaust bia select minut ultra sparc partit greedi bia select minut cognit bia approach featur select tabl bia select summari rel pronoun resolut result boldfac respect baselin improv respect combin recenc baselin individu bia recenc combin recenc recenc combin greedi bia select exhaust bia select discuss summari result tabl summar result appli cognit bia approach featur set select rel pronoun task tabl show perform learn algorithm steadili relev ad baselin note statist gain recenc bia appli isol merg recenc gain perform respect baselin incorpor subject access restrict memori greedi exhaust bia select addit improv perform featur set select method obtain improv spite fairli small data set size larger data expect perform infrequ occur type perform markedli combin recenc rel pronoun fairli long averag number featur case averag number featur case entir data set restrict memori bia prune featur distant constitu learn algorithm concentr recent phrase case retriev improv combin recenc due perform syntact ambigu rel pronoun construct section exampl govern publicli show recenc women prison recommend person carri abduct select victim politician member recenc distinguish show syntact ambigu case semant hard person provid consist inform presenc collect mass noun clair cardi read rel order disambigu rel pronoun current case includ featur phrase rel make handl type ambigu section show greedi exhaust search conjunct cross valid automat bia select experi greedi bia select adequ interact cognit limit exhaust bia select perform slightli small gain perform greedi select worth run time addit data set concentr evalu cognit bia approach featur set select singl data set cognit show approach task rel pronoun resolut appli approach addit languag learn task make cognit learn algorithm handl unknown word addit data set correspond lexic tag task address problem encount natur languag process system reach unknown word system lexicon context unknown word nlp system predict word part speech gener specif semant class word unknown word nlp system encount sentenc inform extract gener concern children kill terrorist kill gener system part speech noun modifi semant featur system determin lexic tag data set case rel pronoun data instanc creat automat circu parser repres context system encount unknown word randomli select sentenc muc busi joint ventur corpu figur featur case featur featur describ local context test word morpholog unknown word part speech semant class inform extract concept actual lexic item cognit bia approach featur select phrase featur toyota motor corp set joint ventur firm class entiti nil figur baselin instanc lexic tag case motor corp set joint ventur firm electr corp entri shown word window center unknown word data part speech concept type correspond entri semant class hierarchi defin joint ventur domain hierarchi experi gener semant featur specif semant featur remain featur lexic tag case encod semant featur inform extract concept major syntact constitu direct recent recogn time unknown word encount case includ class valu predict unknown word gener specif semant featur case rel pronoun case reflect syntact semant inform circu process text specif handl unknown treat predict task independ featur lexic tag task similar rel pronoun resolut main encod richer descript individu lexic item close proxim unknown word rel pronoun concentr reason current task lexic task structur attach featur lexic tag data set repres neighbor clair cardi recenc bia discard irrelev featur nlp system easili includ featur phrase recogn rel pronoun reli learn algorithm discard irrelev featur preprocess step inflat perform baselin discard restrict memori bia data respond readili number cognit good test bia select recogn relev problem hand semant prime syntact show approach support varieti cognit defin addit lexic tag task semant prime semant prime cognit inform peopl tend respond quickli word semant relat entiti involv interpret process system implement semant prime weight semant featur baselin gener specif semant class word implement encourag case retriev algorithm match semant ignor problem determin entiti pertin current point process reli cognit defin syntact prime weight featur primarili syntact issu part speech word syntact categori result lexic tag task experi investig recenc subject restrict semant syntact prime lexic tag task label bia applic data featur label respect unknown word appli recenc weight assum featur word window recent constitu featur assum recenc term unknown word featur featur token immedi unknown word featur token immedi unknown word featur token unknown word featur token unknown word token incur weight recenc bia reflect fact lexic task time delai ms cognit bia approach featur select tabl incorpor individu cognit lexic tag task boldfac entri respect baselin perform shown cognit bia part speech gener semant specif semant baselin system test class class frequent tag weight random select baselin recenc weight semant prime syntact prime subject access restrict memori recenc max wt subject wt semant wt restrict limit syntact wt time subsequ token begin process experi emploi cross valid individu result lexic tag task cognit isol summar tabl tabl show result paramet set bia tabl show perform default heurist algorithm baselin case includ cognit default heurist select frequent occur class heurist perform weight random select base class frequenc unlik rel pronoun data baselin perform significantli default heurist approach work significantli outperform baselin tag rel pronoun recenc bia perform baselin system case data set spite fact baselin recent item cognit bia boost perform syntact predict significantli perform data set lexic tag task clair cardi tabl combin cognit lexic tag task boldfac entri respect baselin task baselin result greedi select exhaust select individu bia result lexic tag task linguist point recenc weight correspond give preferenti statu featur lexic item closest unknown word consist success lexic tag approach classifi token base inform immedi token combin cognit tabl show result combin bia lexic tag task exhaust greedi bia select algorithm run full nest cross valid exhaust bia select larger data set fairli larg number paramet set feasibl limit number paramet set chose limit number partit cross valid figur run time outer partit close hour exhaust select minut greedi select result tabl obtain ten outer partit partit result bia combin includ random compon restrict memori averag run tabl show bia select algorithm provid perform respect baselin combin significantli improv perform recenc bia tabl show greedi bia select algorithm recenc recenc bia select lexic tag task case ten partit semant tag task suggest featur neighbor word critic lexic tag task restrict memori bia select semant tag task bia discard featur major constitu sentenc featur semant tag task gener semant tag prefer smaller context detail predict requir specif semant requir extend recenc weight context maximum weight cognit bia approach featur select tabl summari greedi bia select result lexic tag learn show select number partit bia combin select natur languag task select number partit bia recenc weight max wt bia restrict memori limit bia recenc weight max wt bia bia recenc weight max wt bia recenc max wt semant wt syntact wt subject wt restrict limit analysi exhaust bia select result show similar recenc bia plai promin select recenc weight focu window gener semant class select recenc weight entir context specif semant class restrict memori bia select slightli higher memori limit chosen greedi algorithm major bia select exhaust approach includ greedi approach result linguist knowledg sourc bia recenc bia import bia lexic tag task find greedi bia select algorithm select exhaust bia select algorithm relat work previou work address role machin learn algorithm recent interest autom method evalu select overview articl special issu journal gordon view bia select search space learn work clair cardi cognit process limit type prior knowledg guid select bia learn algorithm cognit set primit defin space induct approach bia select greedi exhaust search conjunct cross valid procedur order search bia space relat provost buchanan techniqu build induct polici build strategi bia select cognit bia approach featur set select make cognit add structur bia restrict memori bia limit number featur consid learn cognit bia approach guid greedi search algorithm increment combin cognit bia approach featur set select construct learn theori multipl bia select algorithm combin produc individu bia merg procedur section work present innov sourc inspir type cognit prefer previou work featur set select reli greedi search algorithm xu al caruana freitag john al skalak cross valid maron moor work similar spirit search occur featur smaller space cognit search cross valid directli select relev subset featur select cognit direct featur select featur weight cognit bia approach featur weight spirit gener method featur weight winnow focu comparison method learn algorithm learn previou work make cognit guid featur select algorithm approach visual global method comput singl weight vector case extrem local method comput weight vector pair train test case number local weight scheme featur weight vari instanc instanc featur featur metric stanfil waltz earli machin learn algorithm assign weight featur aha goldston associ weight vector train case combin global local comput featur weight greater similar test case train greater emphasi train case weight global weight learn algorithm featur relev vari train instanc rc algorithm domingo algorithm cognit bia approach featur select cluster method perform featur select assign continu featur weight learn system implement local featur weight scheme weight vari class level class distribut weight method comput weight vector class set train case statist properti subset data al featur import assign high weight featur highli correl class classifi calcul weight vector class attempt learn featur weight cycl train instanc adjust valu weight strengthen featur valu match instanc weaken valu match instanc class method comput weight produc similar metric test case recent investig featur weight base inform gain improv perform minor class instanc hasti tibshirani friedman comput metric reli discrimin analysi atkeson al creat similar metric test regress classif cognit bia approach featur select featur weight creat similar metric global similar metric depend data set rel pronoun method comput similar measur base test select appli test case perform featur select featur weight true lexic tag task data case number type attribut result similar metric deriv cognit bia approach global featur select weight case retriev test examin local similar metric base knowledg approach background knowledg form gener cognit process limit guid featur set select featur weight fisher cobweb similar method introduc cobweb conceptu cluster system psycholog similar metric guid concept format creat classif tree train case termin base predict store case system probabilist weight sort test assign weight path tree appli similar metric test case learn approach languag learn daeleman make cognit context window select case clair cardi cognit system implicit previou learn nlp requir design featur set natur languag learn problem approach mark step nlp system underli instanc linguist knowledg acquisit task detail discuss relat work cardi summari research present shown cognit process limit serv sourc bia guid featur set select learner concentr collect learn task natur languag process explor cognit subject short term memori semant syntact prime shown cognit automat explicitli incorpor instanc natur languag learn task introduc algorithm cognit bia select algorithm combin cross valid greedi exhaust search select relev paramet set data set test select bia significantli outperform baselin default heurist data set perform significantli inform featur set select method perform varieti natur languag learn data set experi learn algorithm determin rel pronoun learn algorithm improv relev cognit incorpor instanc lexic tag select improv baselin spite fact number irrelev featur discard baselin preprocess step found greedi bia select work data set spite algorithm limit abil handl bia interact evid set compat cognit select learn task evalu cognit bia combin addit issu address futur work featur set select method test larger data set larger lexic tag data reduc varianc perform partit possibl perform individu bia select algorithm import line investig develop method identifi cognit bia interact priori bia combin test test approach lexic structur task natur languag process domain cognit bia approach featur select expect cognit bia approach featur set select work learn task applic cognit prefer make exampl includ speech understand investig imag classif focu color design adapt focu restrict memori recenc color prefer plai hope appli method present learn problem domain futur work investig cognit aid featur weight algorithm learn creat counterpart induct learn algorithm present elimin linguist annot train text supervisori inform acknowledg david skalak discuss comment earlier draft paper anonym review editor david aha construct comment suggest research report support part nsf grant note semant quantiti subset featur specif domain train instanc extract set requir text domain weight chosen base preliminari test valu test subsequ section compar accuraci cross violat train set size assumpt test cross valid result baselin singl bia experi nlp task chosen cross valid valu gener semant specif semant refer toler attribut learn algorithm intern journal concept learn flexibl weight fourteenth annual confer cognit scienc societi lawrenc erlbaum associ learn algorithm machin clair cardi local weight learn artifici intellig percept commun pergamon domain knowledg influenc similar judgement reason workshop san morgan kaufmann tree improv learn tenth intern confer machin learn san morgan kaufmann empir method inform extract ai integr learn cognit machin learn natur languag journal theoret artifici intellig improv minor class predict featur weight fourteenth intern confer machin learn san morgan kaufmann advantag mention spanish bulletin greedi attribut select eleventh intern confer machin learn san morgan kaufmann nearest neighbor pattern classif ieee transact inform trade mip memori knowledg engin commun restrict late closur strategi spanish special issu learn natur languag journal theoret artifici intellig van den forget except harm languag learn machin individu work memori read journal verbal learn verbal individu integr inform sentenc journal featur select lazi learner artifici intellig find structur time cognit knowledg acquisit increment conceptu cluster machin model cognit bia approach featur select flexibl metric nearest neighbor classif unpublish manuscript build access clausal advantag mention versu advantag recenc journal memori recenc prefer twelfth annual confer cognit scienc societi lawrenc erlbaum associ attach evid english spanish sixth annual sentenc process confer massachusett evalu select machin learn machin discrimin adapt nearest neighbor classif unpublish manuscript examin local vari weight nearest neighbor algorithm intern confer reason springer irrelev featur subset select problem eleventh intern confer machin learn san morgan kaufmann principl surfac structur natur languag individu syntact role work memori journal memori element machin learn san morgan kaufmann sentenc exploit world advanc connectionist neural comput theori ablex publish descript circu system messag understand confer san morgan kaufmann learn quickli irrelev attribut algorithm machin race model select lazi learner artifici intellig magic number minu limit capac process inform psycholog machin learn messag understand confer san morgan kaufmann clair cardi messag understand confer san morgan kaufmann matur constraint languag learn cognit process sentenc phd massachusett institut ma induct pragmat bia select machin program machin learn san morgan kaufmann overfit avoid bia machin repres case knowledg sourc appli local similar metric fourteenth annual confer cognit scienc societi lawrenc erlbaum associ prototyp featur select sampl random mutat hill climb algorithm eleventh intern confer machin learn san morgan kaufmann reason commun lexic access sentenc context journal verbal learn verbal review empir evalu method class lazi learn algorithm artifici intellig mit encyclopedia cognit scienc mit press strategi featur select ninth intern confer pattern recognit ieee comput societi press receiv septemb accept octob final manuscript august 
IR/1548 5 machin fl kluwer academ boston manufactur netherland adapt retriev intern local context scale web filippo menczer richard belew comput scienc engin depart california san diego la ca usa jaim yime william cohen abstract paper machin learn abstract spring ecolog evolutionari adapt local select queri intern signal outlin number experi point feasibl perform method gener class graph environ describ method appli intellig retriev inform distribut network environ paper discuss distribut evolutionari algorithm construct popul adapt web agent infospid search inform relev hyperlink autonom intellig fashion adapt spatial tempor regular local context result suggest infospid complement current search engin technolog start search engin stop engin provid global start base statist featur search space infospid topolog featur guid subsequ search show approach overcom limit current state deal problem local distribut inform world wide evolutionari local reinforc neural relev linkag relev select queri graph adapt agent introduct inform environ tool assist user wai scale growth environ adapt person prefer user user condit situat unlik face ecolog organ adapt natur environ natur agent adapt topolog intern behavior evolut learn signal cue perceiv lead featur correl fit intern ant associ pheromon food intern ant behavior behavior implement ant central ant resort random implement absenc signal capabl ant coloni intern distribut integr extern driven endogen behavior highli desir properti agent carri task complex artifici environ paper machin learn abstract spring ecolog applic intellig retriev inform distribut network environ section discuss limit current state section suggest wai limit overcom emploi adapt agent idea test graph abstract inform discoveri section illustr section context implement adapt web retriev agent system evalu section section discuss main machin learn contribut work domain network inform contrast approach address similar problem section conclud outlin direct futur research state art exploit proven techniqu inform search engin growth web provid user assist attempt locat retriev inform web search engin continu grow servic offer success multipl popular model search engin draw effici process inform collect document produc amort cost process larg number queri access index index basic invert file map word collect set document word addit process involv step remov conflat word stem word weight scheme sourc search engin opinion limit fact assum collect case earlier inform retriev system case collect highli document move time index reduc web continu updat crawler exhaust visit period revisit web page time index inaccur stale inform recent delet move incomplet miss inform recent ad chang compound huge size hinder search engin capabl satisfi user queri user face larg hit low retriev page relev respect stale inform factor make user invest time manual neighborhood subset hit list scenario requir attent user suggest queri imagin user profil track inform relev interest user subscrib mail digest human form high price case singl resourc perfectli match user multipl resourc cover wider set interest user user profil requir consciou effort subscrib inform sourc problem partial address advent push limit choic channel broad scope question extend tradit inform retriev techniqu address answer suggest popul autonom adapt inform agent effect deal problem autom work user crawler autom inform gather task build search engin endow similar agent limit amount adapt capabl work behalf user subsect discuss wai agent improv current state art linkag topolog index process build statist topolog document space vector document queri view vector larg featur space word correspond dimens document document relev respect angl respect vector small search engin show similar document effect creat fly topolog base word statist model user immedi make assumpt content retriev fact word network inform environ addit structur provid user cue focu linkag inform basi hypertext markup languag web submit search engin queri document link space store inform scale exponenti linkag inform lost construct exploit fact navig document document tabl measur ten queri submit lyco onlin minimum score paramet lyco full http refer consid articl refer eb queri repres boolean search queri color blind photographi einstein bach cinema internet evolut select red wine link argu linkag topolog spatial structur document number link preciou asset web unstructur portion author tend cluster document relat topic point link linkag topolog browser expect link provid guidanc wast time quantifi notion ad linkag topolog conjectur captur extent linkag topolog relev respect imagin user agent random walk strategi defin condit probabl random link current document lead relev current document relev call relev autocorrel defin probabl document fraction relev document call gener random probabl find relev document jr probabl current document relev linkag topolog random lead relev document higher random frequenc order occur inequ simplifi express linkag topolog ad quantiti realiti measur queri coupl search engin relev autocorrel statist collect count fraction document relev point back document set gener statist collect normal size relev set size collect gross base assumpt set return search engin correspond relev set conjectur ad linkag topolog confirm larg valu shown tabl note document structur linkag topolog driven smaller size collect valu significantli posit independ evid linkag topolog found bibliometr studi web approach web page categor discoveri link constitut cue exploit autonom agent user agent exploit linkag inform confid unreason task autonom agent note inform encod word linkag topolog arguabl complimentari construct manual point page reflect author attempt relat write word topolog consequ word vocabulari choic made page make agent perceptu sensit word topolog featur capabl act link expect find interest relationship manual linkag web author word scalabl discuss previou scalabl major issu limit effect search engin factor contribut problem larg size rapid growth highli dynam natur order index crawler revisit document delet crawler exhaust visit document index complet crawler behavior load document examin period heurist estim frequent document chang accuraci statist highli volatil network load scale number document web time scale time visit document longer stale inform index number queri answer search engin unit amort cost queri scale agent search web scale problem search current environ run stale inform effici search engin amort cost search queri assum user cope longer wait queri search engin answer impact search agent network scale make index free suffici network resourc complet absorb impact search search engin factor inform index correspondingli stale maintain constant amort cost refin result queri search amount network resourc scale ffl ffl imagin visit web page accept ffl achiev impact network load satisfi condit assum number queri pose constant time current growth web assur condit met simpl shift network impact search ffl achiev balanc network load search engin crawler agent context sampl includ document index web search depend heavili share context document author make intend audienc document medium academ typic reader understand intend thing web huge audienc bring share author intend context vagu linguist concern concret manifest global word frequenc statist collect web search engin util index discrimin relev irrelev muddi averag applic distinct word mean situat reli local coher keyword distribut exploit structur proxim agent intern featur describ current document discrimin relev page agent page attribut weight word depend queri satisfi music sport agent situat environ local context analyz word mean situat approach word surround link document provid agent valuabl inform evalu link guid path statist approach action select index construct knowledg queri user pose rank scheme gener good specif queri user person agent adapt user chang time intern user prefer respect word rel import term local person aspect context usag search engin provid person agent captur featur local person context adapt improv search perform search agent idea discuss previou section agent framework goal address limit search engin respect adapt algorithm advantag statist linkag topolog distribut inform environ argu agent person browser approach achiev properti base idea system problem simpler address simpl agent perform simpl oper philosophi drive view agent document document make autonom link adjust strategi local context person prefer user dynam bia search area framework individu agent popul adapt individu learn solut reinforc captur global featur search space user heterogen solut complic intern model model make learn problem difficult adapt evolutionari system prone prematur converg genet evolv solut reflect inappropri due individu agent incap learn life reason motiv hybrid genet algorithm local search reflect gener problem machin learn techniqu environ larg featur space local select long agent live global made agent die order bia search answer question satisfactori algorithm motiv model ecolog organ initi popul energi foreach aliv process sensori input produc motor action updat energi state option learn reinforc reproduc split energi offspr die replenish resourc figur pseudocod evolutionari algorithm base local select ing natur environ provid wai remain agnost question algorithm illustr figur step agent popul initi random search behavior initi reservoir algorithm implement agent call random order avoid spuriou execut parallel step depend sensori motor apparatu agent gener assum action incur work possibl energi intern energi reservoir updat step interact environ result cue reward penalti signal exploit reinforc learn step agent adapt lifetim step central point algorithm differ evolutionari algorithm agent kill select reproduct energi conserv event select threshold constant independ rest popul select local fact reduc commun agent process minimum import consequ agent compet situat portion environ share resourc environ drive competit consequ select pressur process strictli local central made long agent frequent die search directli environ size popul ag distribut determin emerg carri capac environ determin energi replenish resourc occur step replenish independ popul adapt make algorithm amen cover optim good solut repres popul standard converg criteria true respect search bia exploit resourc locat singl resourc multimod optim inform goal locat relev sourc inform algorithm paralleliz lend ideal distribut case agent execut remot server server comput algorithm achiev speedup proport size popul local select disadvantag limit imagin popul agent execut code remot server distribut inform data client machin page access typic situat central page cach commun overhead synchron parallel speedup achiev case hinder scenario feasibl distribut implement evolutionari algorithm base local select requir environ data structur natur agent environ local interact advantag previou experi applic trivial case physic state sensor provid situat agent histori inform object resourc softwar agent provid execut environ capabl access local storag capabl requir solut dynam resourc servic payment transact local select algorithm immedi appli arbitrari problem problem space lend data structur mark environ continu function optim impli discret search space unclear map resourc situat map arbitrari feasibl local data structur difficult isomorph problem resourc model appli successfulli associ resourc partial solut combinatori optim problem requir decomposit properti problem unknown possess problem network inform search pose challeng local select algorithm limit local select weak select pressur energi level agent lifetim depict figur agent reproduc time give half energi offspr final agent run energi di shortli time select event rare energi time figur typic energi dynam agent lifetim higher curv plot level accumul energi function result instantan energi plot lower curv select threshold except reproduct energi level integr lower curv long energi level fluctuat select pressur issu address section intern agent interact sensor pick signal pure inform correl perform agent agent access reinforc signal environ assess appropri signal reward penalti signal adjust agent behavior life basi reinforc learn framework agent perform complex network inform access signal problem signal identifi cue correl perform agent discrimin good bad behavior discuss section inform agent deal problem assum agent learn posit correl perform imagin energi chang comput step algorithm figur instantan energi chang correspond time deriv agent energi plot figur signal continu sign repres perfect cue reinforc learn scheme adjust behavior agent action perceiv lead reward action perceiv lead penalti discourag consist reward time figur warrant lead earlier prevail penalti incur time lead delai death effect intern signal reinforc learn lifetim agent obviat problem weak local select pressur agent pick cue detect time scale evolutionari adapt select base signal integr longer time averag produc agent behavior robust face cue evolut local select intern reinforc learn face process inform environ improv reinforc unit agent signal sampl unit popul signal integr time view local select extens reinforc learn individu reinforc evolut form adapt integr cover tempor spatial scale cue detect search random graph section outlin number experi point feasibl perform local select intern abstract class graph environ problem broadli search larg graph sublinear time imagin larg node payoff popul agent visit graph agent edg goal maxim collect payoff visit time visit fraction node graph framework studi theoret case problem find optim path unknown node weight graph agent heurist search graph problem interest typic graph agent charg cost edg evalu node payoff issu distribut algorithm central problem class graph search task gener reduc interest special case node model hypertext edg payoff measur problem network inform explor search strategi simul inform model relev graph model environ agent posit move reach goal typic task situat robot instanc graph search node assign payoff uniform probabl distribut unit interv link annot real compon idea properli guid agent predict payoff node base featur link point node make agent genotyp neural net vector weight wn agent receiv outgo link node link featur vector algorithm figur neural net comput wn predict payoff node point final agent link pick stochast selector link current probabl distribut fi paramet compon agent genotyp describ import attribut link predict exist construct optim weight vector neural net predict payoff accuraci agent genotyp follow link achiev optim fit payoff energet benefit action payoff newli visit provid visit agent node track environ node yield energi constant energi cost charg node visit smaller cost charg visit prevent endless path visit node reproduct agent genotyp clone mutat obtain offspr genotyp fi weight mutat addit uniform constraint fi studi effect local select isol recombin oper appli random graph gener number distinct accuraci minimum achiev error payoff predict gener equal densiti payoff threshold section carri capac section tabl graph search problem multimod number cluster relev node distinct optim weight irrelev background optim weight vector autocorrel defin condit probabl relev node link node cluster cf relev autocorrel section graph construct experi section averag featur construct accuraci local versu global select graph environ compar local global select binari determinist tournament select chosen global scheme comparison natur step algorithm modifi tournament select energi level randomli chosen member popul place reproduct death algorithm stop node visit figur illustr differ perform typic observ select scheme graph recal level relev node visit plot function time node visit local select popul continu discov constant rate good tournament popul tend converg prematur notic linkag topolog inform fact situat correspond worst perform local select perform slightli random walk random walker expect recal equal fraction node experi repeat wide rang graph shown tabl graph local select significantli consist outperform tournament select improv vari depend graph gener recal fraction node visit ls tournament recal fraction node visit ls tournament recal fraction node visit ls tournament recal fraction node visit ls tournament recal fraction node visit ls tournament recal fraction node visit ls tournament recal fraction node visit ls tournament recal fraction node visit ls tournament figur perform local select tournament select typic random graph valu error bar standard error multipl run algorithm graph paramet ls tournament figur perform select scheme search graph valu ls tournament figur perform graph valu vari densiti good result notic trend condit figur show base case lowest recal level plot record node visit ls tournament figur perform graph valu correl good import agent situat greater consequ determin futur expect local select perform improv figur show case unimod graph intermedi observ perform tournament select yield consist advantag favor local select make problem multimod expect tournament select degrad perform due prematur converg figur illustr trend case high intermedi advantag favor local select predict intern global cue set goal test capabl agent evolv local select algorithm intern global cue signal consid accuraci payoff predict base link potenti accuraci optim evolv agent high optim agent strategi low intern link predict accuraci impli evolv fi valu strategi implement agent stochast selector run ten experi graph averag consist noisi accur fi initi uniform distribut figur correl link predict accuraci evolv fi averag agent popul linear regress shown rang measur node access figur fi valu evolv popul correl accuraci cue posit correl fi coeffici popul successfulli intern cue accuraci agent behavior intern local cue experi graph aim test local cue intern reinforc learn occur lifetim individu agent endow agent capabl adjust neural net algorithm figur algorithm chosen easi implement connectionist framework agent agent neural net natur function approxim agent compar payoff current node predict base featur link visit node perceptron weight adjust delta rule improv accuraci link predictor net instantan energi chang minu reinforc futur valu discount factor learn weight inherit offspr reproduct run experi graph experi disabl fraction node visit ls ls figur effect recal perform figur show improv perform agent base network weight correctli predict payoff intern local featur relev cluster versu irrelev background infospid implement method discuss appli construct popul inform agent infospid system implement test perform person agent web section describ infospid implement discuss detail distribut evolutionari algorithm agent algorithm infospid search inform relev make autonom link follow figur show infospid implement local select algorithm central part system relev feedback user assess relev document visit infospid point relev assess place asynchron respect alter subsequ behavior agent chang energi landscap environ process akin replenish user initi energi foreach aliv agent pick link current document fetch document reinforc signal process option relev feedback user figur pseudocod evolutionari algorithm base local select distribut inform agent environ bia search process overview algorithm high detail subsect user initi list keyword list start form bookmark file step popul initi start document agent document random behavior initi reservoir energi step agent local neighborhood analyz text document situat relev neighbor document point hyperlink current estim base link relev step agent link current document step agent energi updat energi order surviv continu visit document behalf user energi agent reward energi visit document relev function agent evalu relev document document visit assess user assess document visit relev estim mechan implement speed process minim duplic transfer document implement infospid pose cach form commun bottleneck perform distribut agent distribut agent local cach current implement simul perform distribut simpli assum cach set cach size energi agent charg energi cost network load incur transfer document cost function depend transfer latenc document size simplic assum constant cost access constant cost access stationari back pair discourag graph instantan energi step signal agent adapt lifetim adapt process agent modifi behavior base prior learn predict link follow step agent kill select reproduct case offspr recombin type crossov local agent recombin agent resid agent recombin agent popul local crossov distribut infospid minim commun overhead remaind paper offspr provid variat adapt agent evolut step user system relev feedback process option place direct interact agent user assess visit document feedback word document automat assess updat encount word word integ count initi updat time document assess word feedback list maintain global profil word relev user output algorithm flux link rank relev estim modulo relev assess user algorithm stop popul extinct lack relev inform termin user agent architectur figur illustr architectur infospid agent agent interact inform consist actual network collect data local disk relev feedback data cach user interact environ access data local client statu web document suggest make relev assess save local client distribut cach inform environ user agent keyword vector neural net weight link estim selector learn reproduct death document relev estim raw html document html parser word remov stemmer document assess word index weight compact document infospid relev feedback lib file figur architectur infospid access agent subsequ report direct interact user agent infospid prototyp written run unix platform web interfac base librari agent emploi standard inform retriev tool filter word stemmer base porter algorithm agent store effici visit document share cach client machin document repres list link stem keyword cach reach size lru recent replac strategi adapt agent figur highlight central depend infospid system agent describ adapt infospid consist determin behavior agent pass offspr actual mechan genotyp implement search strategi compon agent genotyp consist paramet fi repres agent trust descript page outgo link fi initi fi agent genotyp list initi queri term neural net versatil model adapt standard comput devic figur genotyp vector initi randomli uniform distribut small interv keyword repres agent opinion term discrimin document relev user rest weight repres interact term respect relev neural net input keyword genotyp singl output unit input activ valu network neg possibl neg correl perceiv term relev reason network hyperbol tangent squash input activ valu part system base action select agent perform action select comput relev estim link current document feed agent neural net activ small set keyword sensit input unit neural net receiv weight count frequenc keyword occur vicin link experi report distanc weight function keyword occurr close link question link keyword neural net receiv ith occurr simpl count interven link maximum window size link neural link instanc sum match weight agent neural net figur agent estim link current document link input neural net comput count document word match keyword weight decai distanc link network sum activ unit comput activ jk bia jk incom input lower layer output network activ output process illustr figur repeat link current document agent stochast selector pick link probabl fi fi relev estim feedback link chosen document agent determin energi gain depend document visit document user assess agent receiv energi decai decai factor fl paramet avoid popul due energi relev feedback user provid system relev word feedback list repres profil interest current accur origin queri list estim relev unvisit assess energi tanh frequenc term document normal document weight term base relev feedback extens tf idf document index weight scheme word feedback list initi queri term weight user infospid relev feedback case weight updat ff log fraction cach document ff inertia term weight formula differ tradit tf idf scheme respect aim weight term base describ correl relev emploi algebra term frequenc account neg contribut document term relev comput document frequenc base content cach entir collect hyperbol tangent normal energi rang rang neural net predict agent compar relev current document estim link led connectionist version neural net train predict valu link base local context agent visit document intern gener reinforc signal comput teach max futur discount factor predict link neural net weight updat error learn weight inherit offspr reproduct treatment absenc relev reinforc learn algorithm environ reinforc signal relev feedback alter function modifi environ provid exampl relev infospid integr adapt form relev feedback reproduct offspr clone recombin agent crossov appli keyword subset mate keyword splice offspr keyword vector offspr mutat provid evolutionari algorithm power explor offspr fi fi fi fi fi paramet uniform distribut valu fi clip fi max maintain exploratori behavior neural net mutat ad random fraction weight network connect keyword vector mutat probabl term arg min replac term expect justifi agent perform respect user assess order singl keyword mutat term select probabl distribut ji document birth factor captur local context select word describ document led energi result reproduct factor captur global context set user select word discrimin user prefer learn care adjust neural net weight keyword evolut keyword local mutat crossov implement form select queri base relev feedback local queri adapt time place popul agent embodi heterogen model relev possibl inconsist featur agent small set maintain model remain manag face huge featur search space infospid evalu section report result experi analysi aim evalu perform infospid respond queri search web onlin extend preliminari result tabl infospid paramet descript valu experi report paper paramet descript initi popul size reproduct threshold energi cost document tmax max number page visit queri fi initi fi fi fi mutat rang fi max max fi ae link estim slide window keyword mutat rate layer neural net layer initi neural net weight rang iw neural net weight mutat rate neural net weight mutat rang neural net rate discount factor ff inertia word feedback weight fl decai factor document assess max number word feedback entri jcj tmax max cach size section algorithm paramet mention tabl show valu paramet experi section cost agent visit irrelev document run energi effect cach size consid experi effici cach larg visit cost assess document cach algorithm effect simul behavior distribut infospid search stop popul meet success criterion subsect extinct visit total tmax eb search graph difficulti evalu retriev system stem multipl factor lack rank function document problem address subsect gener difficulti lack queri relev set overcom special chunk web select test environ advantag make readili relev set articl larg number queri hyperlink index hyperlink index figur distribut histogram eb human societi graph subset eb topic roughli tenth collect environ made organ hypertext graph page articl constitut index form connect graph remain node form hierarch topic call figur show statist search graph articl manual classifi hierarchi skill human editor titl node queri access social organ public privat activ allevi human human societi jane foundat peac corp famili welfar disast relief hull index relev set queri figur schemat eb search space actual show titl relev articl relev set depict light grai dark grai remov search space articl classifi categori exhaust knowledg document subject set articl root queri node relev set queri topolog defin agent search make problem navig relev set easi remov search space node relev agent access relev node directli index node illustr schemat figur queri section tabl show statist exampl queri experi queri node depth tree group queri span depth depth minim distanc stare node human societi root relev part search actual shortest path relev articl deeper queri harder infospid gener carri relev set farther start point number tabl queri statist exampl depth queri measur branch privat substant procedur law govern econom transact law commerci transact princip form busi associ state util queri vari depth multipl measur obtain singl queri restart search initi condit seed random number perform evalu quantit analyz behavior infospid algorithm eb corpu infospid access subset document rank collect make imposs standard inform retriev perform analysi method requir rank similar measur control paramet metric lend assess perform retriev system search length defin number irrelev document order list retriev front fraction relev set easili extend method imagin visit document list retriev order visit time rank length refer wait time retriev system mean wait fraction relev set count total number document visit time queri experi discuss search length measur recal level ne relev articl visit compar infospid global search algorithm search annot optim strategi algorithm case web admiss criterion appli suitabl lower bound distanc closest relev node trivial admiss heurist trivial reduc infospid implement search strategi similar bf evolv high valu fi local singl respect global search space assum bf upper bound global search compar perform infospid implement bf mean prioriti queue link estim global sort estim relev point document visit queu order determin prioriti complic due potenti distribut implement sequenti visit document retriev server resid analyz local client machin case cach minim number repeat document request network case bf arbitrarili long unvisit document contribut measur search length expect algorithm implement distribut fashion order make part system requir commun agent share access global differenti cost visit crossov sake comparison disallow includ document visit search length agent request document network load document cach agent execut search length maximum number document visit agent lineag measur experi track number link accumul gener agent start run run experi compar search length achiev bf variant relev feedback relev feedback newli visit document case assess automat gener document visit belong relev set figur show percentag queri successfulli complet algorithm case relev page found tmax limit visit page complet queri infospid run prioriti queue run visit area perform degrad queri depth infospid higher success rate relev feedback afford improv deepest queri due evolutionari reinforc provid relev agent move direct receiv energi boost greater chanc reach relev cluster figur plot search length versu queri depth search length averag complet queri plot infospid search length depth global heurist show depend observ depth rang bf outperform infospid deep queri practic case expect search start desir page search engin seed initi popul infospid realist queri lower queri depth complet queri prioriti queue infospid infospid figur percentag queri successfulli complet bf version infospid queri depth prioriti queue infospid infospid figur perform infospid relev versu bf error bar correspond standard error tabl default initi word feedback list queri organ public privat servic social human activ allevi distribut natur infospid result significantli shorter search time requir bf figur show user relev feedback discoveri relev point agent direct improv statist deep queri due fewer measur micro analysi illustr goal set adapt inform agent achiev close typic agent adapt singl search queri organ public privat activ allevi human remov stop word result keyword vector shown tabl simplifi analysi run simpl perceptron repres agent agent complet vector keyword vector weight keyword bia fi paramet ran search popul visit provid popul relev feedback page adapt context infospid intern word featur spatial local linkag adapt spatial context answer question born time place born time sequenti execut popul collect visit page born time effect contemporari tempor context global inform result relev assess time partial shown tabl tabl part word feedback list weight time tabl star mark term present origin queri note correspond articl word remov correspond ordin number centuri rank foundat red mission social cross hull hou organ servic activ public human privat allevi tabl keyword vector agent organ organ public public servic servic social social human human activ activ hull hull tabl frequent term document agent born word frequenc normal total number word document da db hou hull poor social centuri american tabl origin queri word displac top posit replac term privat allevi low foundat red highest correl relev feedback time bs keyword vector shown tabl evolut lead queri term lost genotyp gener parent lost allevi mutat favor hull privat mutat gener allevi privat replac hull mutat crossov keyword vector demonstr featur correl relev intern agent behavior differ attribut evolutionari adapt spatial local context born document da db word frequenc distribut partli shown tabl repres place frequent term repres place tabl part word feedback list weight time rank servic social organ foundat nation public activ hull privat human allevi american frequent term intern situat agent suit respect spatial context adapt context adapt tempor dimens infospid intern featur captur tempor context answer question born place tabl time born time tempor context global inform result relev assess time partial shown tabl bs tempor context tabl make observ tabl relev document origin queri term gain top posit rel posit privat surpass human word lost presenc bs cs birth page db hull maintain strong term made american repres db present word feedback list time tabl keyword vector agent tabl organ organ public public privat servic servic social social human american activ activ hull hull tabl show differ agent differ reflect time agent born correl relev true born intern term consist chang tempor context time cs birth american small posit global presenc term db pick imposs due term absenc earlier word feedback list evolv intern agent adapt respect tempor chang time user relev agent effect achiev person learn experi question infospid intern local context smaller spatial scale shorter time agent initi popul live end run success born time search path life experi tabl show weight vector ds es neural net end run weight adapt agent estim document relev link clear strategi learn paid special attent public es predict presenc public demonstr local context experienc live diverg respect word reinforc agent intern neural net perceiv correl featur relev tabl learn neural net weight agent organ public privat servic social human activ allevi bia relat work distinct machin learn inform retriev issu blur reflect approach section discuss main connect contribut area categor issu complet discuss mean complet machin learn contribut evolutionari algorithm local select gener adapt paradigm distribut agent shown approach suitabl domain combinatori optim local select proven success multimod optim problem requir heterogen cover search space converg perceiv global optimum evolutionari algorithm commun behavior refer nich select notabl select variat explicitli aim nich crowd fit share method select alter account measur similar lead popul select time complex individu popul size requir maintain cover nich grow rapidli number nich local select natur enforc mainten popul implicitli nich commun overhead afford effici central distribut task local select reinforc learn agent intern featur fashion infospid integr adapt learn seamless global optim advantag signal spatial tempor scale plenti data noisi inconsist compar manual construct relev adapt algorithm advantag cheap realist actual search environ connectionist model infospid learn estim link relev reminisc neural network learn probabl distribut text retriev logist regress relev assess user activ learn advantag shown improv perform retriev system system integr adapt driven relev feedback model local select relev feedback interact asynchron agent relev feedback selfish process user standpoint agent modifi reward improv model relev perform distribut text collect typic massiv data set challeng machin learn techniqu due huge featur space infospid deal reduct situat agent intern word maxim correl object local context model featur select size learn problem manag individu popul larger set heterogen featur inform retriev contribut learn appli extens inform automat classif cluster differ infospid approach adapt driven local singl agent intend learn global optim document link relev estim strategi heterogen network inform retriev task view averag asset improv perform local level local adapt system relev feedback mainstream text retriev closest learn estim word probabl distribut main differ infospid cue provid user assess conjunct local agent wast limit resourc pai attent word current search user word lot featur select inform retriev queri case word featur intern agent adapt mutat user constitut highli dynam heterogen queri necessarili forev infospid select perform queri respect local context popul deal scalabl agent collect select agent learn task harder idea incorpor infospid framework step address challeng pose text classif machin extend inform retriev deal document user heterogen collect web search make classif problem simpler class irrelev respect current relev class heterogen user class membership chang time user shift interest aspect asset construct retriev system relat work idea decentr process divid task local perform set central perform set suggest earli dai web harvest project linkag topolog consid context motiv link enhanc relev judgment incorpor queri formul improv search fish search search system time infospid inspir idea artifici life fish search base popul search agent web driven intern gener energi measur base relev estim popul central cach effici algorithm extend distribut agent intern local context due mixtur depth breadth cutoff level difficulti fish search approach determin cutoff level possibl result search behavior webwatch agent learn mimic user shoulder perform search user suggest link follow letizia autonom interfac agent assist user web make suggest page interest user agent learn predict object function track user prefer unlik webwatch letizia singl importantli user order autonom search fab amalthaea adapt filter system inspir genet artifici market model term weight relev feedback adapt match set discoveri agent search engin set user profil system learn divid problem simpler deal heterogen dynam profil queri system perform activ autonom improv scale limit index exploit mechan infospid decid stop run energi admittedli crude approach solid fact suggest agent approxim optim stop criterion idea weigh diminish return continu accumul visit futur option evalu market summari result suggest inform agent complement current index technolog start search engin stop engin provid global start base statist featur search space agent topolog featur guid subsequ search shown approach extend current state art deal problem local evalu infospid collect perform encourag support popul locat relev document larg distribut corpora faster advantag distribut model implement shown infospid outperform exhaust search domain order perform receiv boost synergi individu learn relev feedback micro analysi enabl determin singl agent fact intern import local featur environ intern collect ecolog captur heterogen snapshot featur correl user relev agent strategi evolv time chang agent agent agent depend tempor spatial context part environ experienc futur direct earli experi interest behavior agent carefulli control structur corpu full expect agent interact real web remain demonstr shown diverg featur agent success topic area encyclopedia real evolutionari method adapt wider variat found web media expect role type sensit wide vari user effect search dispar corpora extens evalu approach actual web verifi goal make problem evalu open environ high prioriti aspect model explor greater detail futur effect cach cach size perform consid sequenti infospid role distribut cach studi distribut implement weight scheme suggest idf local relev estim mechan inappropri global featur term weight mutat process modifi queri absenc relev base sole correl easili achiev replac factor ji ji probabl distribut keyword mutat section form direct agent interact consid crossov agent reproduct recombin intern nearbi situat server intern experi relev proxim model interact agent worth explor domain agent learn agent agent commun languag exampl activ research area opposit argu distribut agent suit perform task distribut environ point view agent interact minimum call studi interact implicit current central repositori resourc relev feedback begun explor issu distribut agent mobil agent execut server machin inform current agent code agent intellig select inform back user inform provid opinion interest trade cpu cycl improv network bandwidth secur languag protocol trust autonom agent agent research provid system technolog thrust make mobil agent feasibl integr search search engin put test hybrid system construct search engin provid agent good stare base statist topolog search space crucial shown section perform distribut search degrad distanc start point relev cluster person agent continu search adapt user current context popul autonom agent search engin understand mutual benefit approach potenti synergi ensu acknowledg author grate make cd collect experi data part infospid code free softwar massachusett institut matthia dan sourc make softwar gnu gener public infospid project origin collabor wolfram mong wrote graphic interfac earli demo david lewi provid annot bibliographi machin learn inform retriev member cognit comput scienc research group cse depart ucsd discuss suggest note similar argument appli web inform retriev system base search engin queri make conserv assumpt obtain lower bound ad linkag topolog conjectur cluster hypothesi hypertext deriv definit associ retriev set view relev set queri alta time write estim conserv growth rate doubl condit met year list typic obtain consult search engin altern distribut implement crossov agent situat server gener statist account remov relev node length prioriti queue set equal initi size infospid scenario infospid imagin agent carri small local cach refer mitchel learn world wide web aaai spring symposium inform gather distribut ao ga mihaila applic web queri languag proc intern world wide web balabanov adapt web page recommend servic proc intern confer autonom rk belew editor adapt individu evolv model algorithm santa fe institut studi scienc complex addison cm pb mf schwartz scalabl internet resourc research problem approach commun kleinberg automat resourc compil analyz hyperlink structur text proc intern world wide web ws cooper expect search singl measur retriev effect base weak order action retriev system journal american societi inform de bra post inform retriev world wide make search feasibl proc intern world wide web ka de jong analysi behavior class genet adapt system phd ka de jong decentr select algorithm proc intern confer genet digit equip corpor fox lexic analysi stop list inform data structur algorithm wb frake stem algorithm inform data structur algorithm de goldberg richardson genet algorithm share multimod function optim proc intern confer genet harik find multimod solut restrict tournament select proc intern confer genet harman relev feedback queri modif techniqu inform data structur algorithm hart rk belew optim genet algorithm hybrid local search adapt individu evolv model algorithm addison yannakaki search fix graph proc intern colloquium languag page rr larson bibliometr world wide exploratori analysi intellectu structur cyberspac proc annual asi dd lewi activ relev feedback inform retriev aaai fall symposium activ dd lewi inform retriev statist larg data set proc nrc massiv data set dd lewi challeng machin learn text classif proc annual confer comput learn lieberman autonom interfac agent proc acm confer comput human lin reactiv agent base reinforc teach machin rm ba huberman surf real option proc comput econom lyco sw popul size share method foundat genet algorithm sw comparison parallel sequenti nich method proc intern confer genet menczer adapt retriev agent heurist neighborhood inform discoveri proc intern confer machin menczer rk belew complex environ complex behavior adapt menczer rk belew latent energi environ adapt individu evolv model algorithm addison menczer rk belew adapt inform agent distribut textual environ proc intern confer autonom menczer rk belew local select proc annual confer evolutionari san rk belew endogen fit paradigm adapt inform agent cikm workshop intellig inform mitchel machin chapter mouka evolv inform filter solut amalthaea proc intern confer autonom pack ml aw moor reinforc survei journal artifici intellig shneiderman navig design toolbox commun se robertson spark jone relev weight search term journal american societi inform page de ge rj william learn intern error propag de rumelhart jl parallel distribut explor microstructur volum bradford book kotz transport inform agent proc intern confer autonom salton bucklei improv retriev perform relev feedback journal american societi inform salton mj mcgill introduct modern inform retriev sparck jone statist interpret term specif applic retriev journal sparck jone experi relev weight search term inform process spertu mine structur inform web proc intern world wide web person commun steier statist semant phrase hierarch context phd comput scienc engin san steier rk belew export statist analysi topic languag casei symposium document analysi inform sutton reinforc learn inform access aaai spring symposium machin learn inform cj van rijsbergen inform retriev edit refer librari version dk hierarch network search engin exploit hypertext cluster proc seventh acm confer 
IR/2052 5 index retriev natur languag ternari express lin submit depart electr engin comput scienc partial ful llment requir master engin electr engin comput scienc massachusett institut technolog februari reserv author grant mit permiss reproduc distribut publicli paper electron copi thesi document part author depart electr engin comput scienc februari certi ed bori katz princip research scientist thesi supervisor accept arthur smith depart graduat student index retriev natur languag ternari express lin submit depart electr engin comput scienc februari partial ful llment requir master engin electr engin comput scienc abstract tradit inform retriev system base paradigm complet captur semant content document imposs current technolog build practic inform access system fulli analyz understand unrestrict natur languag avoid complex natur languag understand construct inform access system capabl process unrestrict larg understand answer natur languag queri high ternari express suitabl structur express inform retriev amen rapid index thesi bori katz princip research scientist acknowledg mom love uncondit support give machin understand memori comput hope bring step closer special import role plai life special shout real program document origin paper bori katz sue submit titl sue suggest valuabl comment content introduct deploi system question answer outlin index ternari express transform rule previou work earli work recent work sophist index scheme index relat system architectur gener relat index ternari express match ternari express transform rule creat testb system evalu test system corpu test run result sampl output discuss recal parser error question type bene ts relat semant similar verb modi cation relat queri involv common word futur work inform extract sophist queri process sophist match algorithm resolut express de nite noun phrase signi canc relat transform rule hybrid system list figur sampl transform rule sampl transform rule architectur sampl output relat search result frog keyword search result frog transform rule list tabl comparison relat keyword index chapter introduct growth inform electron peopl potenti access knowledg massiv repositori inform promin world wide potenti remain due lack ectiv inform access method simpli dicult knowledg overwhelm amount inform exist search servic reduc inform overload compound textual inform exist di erent vari unorgan ideal intellig inform access system combin understand memori comput document present idea architectur bring step closer dream deploi system deploi inform access system web search engin built equat document keyword statist semant content queri return document word similar found queri tradit inform retriev system account plethora relationship languag entiti sentenc sentenc document sum individu match document base keyword content perfect er indic keyword captur rich express natur languag set similar word di erent bird young snake snake young bird mean life life life mean bank river bank river river bank interfac design design interfac design interfac interfac design due inabl keyword captur tradit inform retriev system su er poor user queri accur formul natur languag question bear user expect answer similar brown bear live island part polar bear live coast user keyword search system keyword bear live unrel assum keyword search engin perform morpholog stem fairli standard sentenc keyword mammal bear young live tradit keyword search engin understand document irrelev result return user question answer question answer natur languag inform access mechan human easi rapidli requir special train unlik inform return list potenti relev document user question answer system return inform segment directli answer user natur languag queri question wrote declar answer je erson wrote declar independ document independ hall philadelphia inform state system start rst natur languag system question answer world wide web exampl question answer system includ entri text retriev confer annual confer track evalu latest question answer technolog obviou advantag question answer er tradit inform robust inform access system base natur languag unrestrict domain realist expect time languag simpli complex full ambigu correctli analyz comput numer problem common model render full text understand qa track begun unrestrict domain reach current technolog simpli analyz textual knowledg deriv knowledg base directli simplifi sophist natur languag techniqu appli document signi portion semant content captur intract complex languag simpli ed linguist mean distil ternari simpl power structur amen index retriev system prototyp implement idea ternari express basic unit system creat accur document content compar er modular architectur serv testb natur languag inform retriev techniqu system lai foundat inform access system outlin rest document organ chapter describ ternari structur suitabl natur languag inform retriev chapter describ transform rule mechan handl linguist variat chapter examin previou work natur languag process techniqu appli inform retriev chapter detail system architectur system chapter report evalu compar tradit boolean keyword inform retriev system chapter explor issu evalu conduct previou chapter chapter outlin roadmap futur develop enhanc system chapter brie summar document chapter index ternari express sophist linguist analysi perform natur languag text question techniqu viabl larg scale index scheme keyword basic unit captur semant content document avoid complex natur languag understand ignor complex ambigu construct inform access system capabl process unrestrict understand inform answer natur languag queri high varieti structur rang linguist sophist attempt discuss chapter limit success ternari express suitabl solut ternari express captur syntact relationship entiti text intuit view easili express type possess syntact point ternari express view type binari relat binari branch hypothesi linguist ternari express theoret capabl express relationship element sentenc involv distinct verb sentenc repres ternari express null sleep arbitrari compat linguist theori semant point ternari express view manipul predic logic ternari express highli amen rapid requisit inform retriev system structur tree case adapt dicult index retriev ecient due size complex structur viabl current inform retriev system ternari express suitabl solut depth discuss previou ternari express index retriev row tabl relat relat design handl larg amount er scalabl solut term perform minim develop ort oper ternari express formul sql queri sequenc relat algebra serv underli foundat relat index analyz extract relat entiti natur languag ternari inform access system captur di erenc mean pair sentenc phrase chapter bird young snake bird eat snake young mod snake snake young bird snake eat bird young mod bird mean life mean life sql handl index relat chapter life describ life life mean mean life bank river bank river bank river bank river river bank river mod bank interfac design interfac design design interfac design interfac design interfac interfac design interfac design design interfac abil extract inform access system distinguish di erent statement ternari express index di erenti preposit phrase modi cation compound system notion semant word syntax er crucial clue mean case refer side refer nancial institut proxim river noun phrase pair word di erent river interpret lexic river bank chapter transform rule syntact inform structur improv inform retriev system due abil identifi relationship entiti structur face problem linguist phenomenon similar semant content express di erent surfac form simpli wai thing recal neg ect di erent variant mean explicitli solut normal di erent construct structur set sentenc express mean di erent capit capit citi taiwan linguist variat occur level exampl demonstr syntact variat linguist variat demonstr verb argument altern beth levin er excel treatment english verb class verb argument altern declar shock shock countri declar transform rule provid mechan explicitli equat altern realiz mean level ternari express figur show sampl transform rule handl applic question equat question shock shock noun noun figur sampl transform rule relev instanti rule shock shock gener observ english verb verb class undergo altern verb particip altern shown share semant qualiti transform rule requir handl altern speci appli speci pattern ternari express rule gener verb class associ rule condit met rule verb excit figur borrow start rule bidirect side rule impli side rule noun figur sampl transform rule note transform rule encod semant knowledg element common rule written equat sell action bui action verb argument di erent rule encod transform rule appli pragmat er express framework handl linguist variat order question answer system success adequ linguist larg number rule lexicon classi ed verb argument altern pattern good resourc lack world todai rule gener ecient mechan creat rule issu requir research implement transform rule discuss chapter note rule chapter previou work concept index simpl keyword chapter describ previou attempt appli natur languag process techniqu inform retriev earli work decad katz mechan index retriev natur languag semant relat net relat system capabl understand fragment text retriev analog text segment store knowledg base idea index phrase decad fagan experi index di erent type syntact deriv natur languag deriv statist properti keyword syntact phrase consist word pair form phrase gener analyz output system programm natur languag text analysi tool phrase automat text analyz prepar pro gener phrase le pro fagan work primarili noun phrase attach preposit techniqu handl languag phenomena hyphen experi perform compar ect index syntact phrase word creat statist addit normal singl word term vector space model order convert phrase descriptor standard vector space model word pair morpholog stem concaten singl lexic item deriv element automat analyz text analyz prepar analyz pro prepar pro fagan experi show index phrase result improv baselin term index syntact phrase work suggest natur languag process techniqu ectiv pure statist techniqu limit fagan studi phrase descriptor atom index simpl morpholog stem lexic concaten techniqu simpli es index disregard fundament relationship compon word phrase descriptor simpli cation limit type queri form treat pair lexic atom fagan exact stem algorithm lexic item deriv similar exampl render imposs big pair recent work research explor ect index word pair deriv linguist analysi al approach appli natur languag process techniqu inform retriev involv extract index er pair pair gener tag text text process system base linguist string grammar type pair consid includ head noun head noun head noun main verb head object subject pair merg index term gener techniqu system report demonstr viabil appli linguist analysi inform retriev author experi index word pair deriv er relationship perform improv neglig lack signi author arm potenti index scheme advantag er tradit ir pair express captur semant content natur languag text question answer deriv pair di erenc mean pair repres modi cation pair chapter bank river bank river interfac design design interfac design interfac interfac design type pair ternari creat structur suitabl captur document content inform access sophist index scheme sophist linguist techniqu emploi analyz text inform retriev rang simpl er complic tree structur case frame index linguist tree structur attempt disappoint due inabl handl variat tree structur semant content express di erent syntact poor qualiti natur languag slow natur languag parser smeaton index incorrect tree sourc perform degrad match tree intens full linguist tree deep relat easier match typic simpler tree tree happi peopl issu address transform er potenti solut technolog advanc year interest appli robust parser todai describ peopl happi describ peopl peopl index case frame attempt limit success full semant analysi open research gener domain analysi perform case frame analysi inherit current parser semant analysi requir extens knowledg extrem construct due complex structur case dicult store retriev ternari express index relat previou attempt appli kind relat question answer system perform question answer onlin encyclopedia util boolean keyword base approach guid heurist narrow broaden search paramet limit small number xed phrase type recogn order extract verifi hypothet list inform evalu system seventi rst result return correct correct answer li top ve result time question answer system util semant relat focu work extract semant relat syntact involv addit comput lexicon full semant analysi system settl syntact relat subject small number easili identi semant relat time inform extract techniqu identifi type relat system receiv reciproc rank methodolog score slightli averag score trec particip primarili trec lexicograph similar work instruct compar similar corpora similar task start system languag index ternari express order provid question answer system inspir work start perform full analysi sentenc fulli unrestrict restrict queri sentenc phrase describ complex text start repres content index form knowledg base annot manual start index larg corpora unrestrict text start robust capabl domain knowledg demonstr ternari express natur languag question answer chapter system architectur system framework inform access system implement idea present earlier chapter system design extens exibl hope serv testb futur experi natur languag process techniqu appli question answer architectur system shown figur written perl modul api program extern depend system parser creat handl storag index relat access extern depend ned form architectur interfac question answer form websit view repositori inform segment system index segment tag natur languag similar model start inform segment multimedia content annot natur languag sentenc describ inform question capabl annot analyz convert ternari latin verb web interfac wrapper figur architectur index inform segment annot system design lead greater exibl futur extens segment consid mechan gener annot automat technolog spoken technolog gener relat system emploi parser gener relat corpu text minimalist parser principl essenti constraint er advantag uni grammar major advantag true segment html tag displai properli download translat fast system achiev respect speed perform sacri coverag accuraci handl wide rang phenomena passiv except case displac tree output postprocess wrapper order convert ternari express wrapper tree perform match local tree fragment fragment convert ternari output sentenc figur big bad wolf dark forest sentenc deriv mod mod mod index type modi modi possess predic predic preposit rel suitabl parser extract relat natur languag question answer ti current system handl sentenc presid russia visit presid deriv visit system distinguish di erent determin visit solut problem add indic implement fin det det mod mod wolf subj prep mod det det mod figur sampl output output tree di erent linear graph structur collect tripl parser substitut text analysi provid similar wrapper written index ternari express ternari express extract text index popular open sourc rdbm manag ternari express neatli relat model manipul sql queri oper ternari express neatli translat sql rdbm index store relat vastli simpli es develop relat design handl vast amount intellig indic instant access ternari express store relat index system costli lack robust product qualiti softwar packag sql scalabl solut index retriev problem interfac perl modul connect method industri driver popular lightweight rdbm easili substitut larger load involv larger corpu match ternari express ternari express matcher compon system directli gener answer user queri modul directli interfac inform access system web parser displai tree di erent wrapper rewritten ternari express deriv queri manner corpu text process mediat question substitut wildcard match love deriv love type wildcard produc exact answer ternari express match found element bind answer question matcher emploi simplist inform segment score number relat share queri score sentenc ternari express common queri segment score greater return queri result format output displai top result current match algorithm inadequ chapter line philosophi algorithm replac advanc version written er roadmap futur work transform rule system implement modul transform rule system essenti forward chain consequ rule ternari express system attempt unifi relat gener text transform rule uni cation pattern introduct uni ternari express consequ proper bind instanti transform rule appli time index time retriev rule applic variat corpu text gener transform rule insert index slow index process result larger index speed match retriev time rule applic perform retriev transform rule appli user broaden encompass variat gener reduc index size retriev time util combin frequent occur transform bene rule index size reduc appli infrequ encount transform time retriev addit actual machineri intuit interfac ad rule transform rule written automat compil pattern ternari express rule figur written shock xs xs shock compil de nition pattern ternari express proper free variabl user option review ternari express gener check manual edit rule mechan attach condit rule handl verb creat testb intend demonstr potenti improv current system result index ternari primari establish extens framework futur improv hope system function test techniqu aim build inform access system roadmap futur develop outlin chapter chapter system evalu chapter evalu compar inform access prototyp relat baselin keyword index setup result present test system detail descript test system relat index matcher sentenc corpu relat deriv tree index matcher extract relat user queri sentenc corpu score number relat share queri top score sentenc corpu return user answer system implement function modul transform util evalu keyword index matcher keyword index deriv content word corpu stopword morpholog stem standard invert index system previou chapter built content word serv data structur match retriev matcher extract queri term user question stopword ltere morpholog stem engin attempt match keyword queri logic keyword invert index sentenc rank order number keyword share queri top score sentenc return answer corpu test corpu experi electron version encyclopedia raw articl origin packag sgml signi preprocess requir prepar corpu index encyclopedia approxim articl vari greatli sentenc page entir collect hundr thousand sentenc million total approxim megabyt sgml tag test run twenti question list submit system question attempt ect typic question real user seek knowledg countri japan frog eat countri japan lion capit capit found invent electr largest countri silk largest defeat spanish ant start civil primari test serv formal evalu highlight advantag index retriev text base relat result comparison keyword index relat index tabl keyword index achiev relat index improv gure enorm rise due part test design highlight advantag index chapter addit achiev higher system return fewer result averag compar system result queri result keyword search engin return larger number correct answer relat search engin correct answer queri correct answer discuss result present chapter relat keyword question countri japan frog eat countri japan lion capit capit found invent electr largest countri silk largest defeat spanish ant start civil averag tabl comparison relat keyword index total number sentenc return number sentenc correctli answer question sampl output output keyword index question frog shown figur output question relat index shown figur remov stopword keyword search engin return result keyword frog eat sentenc correctli answer user result answer question eat poor frog predat approach fundament di erenti queri frog predat queri frog prei relat search correctli identi es relat return correct result lter sentenc di erent relat abil captur mean english reason index relat lead higher inform access system index relat obtain higher system util recal system lower answer user queri phrase construct di erent queri problem linguist variat ect fewer correct result return relat compar keyword index chapter explor issu return due complic compar frog frog frog syntact di erent mean frog frog adult frog eat insect small includ spider figur relat search result frog frog eat kind small anim includ small bird bat bat catch sh speci eat small tree bat bow bow ns eat sh eat small eat kind small mammal snake eat insect frog crane crane eat varieti includ plant tiger snake snake eat frog varieti small mammal electr electr eat frog smaller sh young snake eat adult feed small mammal fish sh eat frog frog eat frog egg frog digest system enabl frog develop eat live anim frog adult frog eat insect small includ spider frog unit peopl eat leg green frog snake catch eat frog eat small includ insect king king eat insect eat small mammal eat small small bird lung sh lung sh eat small sh water frog seiz eat small tree frog eat live includ locust eat bird eat bulb eat small mammal snake snake snake eat small mammal spider spider feed spider eat small mice spider enemi spider includ anim eat insect eat tree turtl snap turtl eat small water eat bird figur keyword search result frog chapter discuss reason obtain dramat improv standard keyword index scheme sampl test question design highlight advantag index relat negat impact signi canc research chapter examin result evalu greater detail recal focu work improv possibl recal due problem linguist answer user queri state di erent construct answer properli retriev ternari express repres document content syntact semant sentenc di er syntax consid di erent matcher sentenc encyclopedia answer question feed young small anim interest appli techniqu question answer track trec confer order compar system approach standard corpu relat deriv sentenc di er relat deriv sentenc question correctli answer problem allevi transform rule chapter encod speci rule equat construct construct relat index retriev correct answer origin question formul di erent transform rule code method construct resourc requir task issu requir research eat feed figur transform rule retriev base relat inadvert omit correct answer answer state di erent worth tradeo sampl keyword search engin retriev correct answer relat search addit answer er repeat inform present present irrelev fact answer return relat search engin question silk silk road group ancient trade rout connect china europ keyword search engin addit list famou great silk silk product ow roman empir open protect silk major trade rout carri silk luxuri good china india middl east citi import stop silk rout west marco trader travel silk carri good europ western asia earli system road silk trade rout ran connect china rome europ northern call silk cut china central asia east coast sea black sea silk trade rout connect middl east europ china addit answer return keyword search engin provid extra inform directli relev user queri inform access system util relat carefulli balanc linguist sophist document analysi techniqu robust design greatli ect recal system perform term speed system extract index complex relat strategi carri perform chapter speed addit index retriev accuraci failur correctli identifi overli complex optim point balanc issu warrant studi parser error appli natur languag process techniqu inform retriev improv qualiti question answer system brittl due parser error sentenc corpu long relat extract incorrect sourc error relat build parser incorrectli construct parser wrapper incorrectli convert tree relat posit error relat neg error text encount due complex natur error unavoid approach question answer depend critic qualiti relat parser fail correctli identifi relat identi es wrong match algorithm easili lead recal potenti su er situat allevi util parser return partial analysi text parser encount complex sentenc return partial tree incomplet partial tree accur repres linguist unit noun ternari express deriv text entir sentenc analyz anticip futur develop natur languag process ternari express extract modul system compon easili replac modul advanc parser question type bene ts relat corpu size bene ts index relat index keyword pronounc relat index ectiv lter sentenc correct wrong type normal inform retriev techniqu perform select larger probabl sentenc correct wrong standard keyword retriev system perform experi requir quantit assess impact parser error note encyclopedia small size class question relat index er advantag semant similar verb verb subcategor argument semant class system verb argument semant type import reorder verb argument drastic alter mean keyword index su er poor queri requir identi cation relat center john issu presid russia critic keyword search engin distinguish question queri keyword content keyword approach unabl determin critic modi cation relat keyword index system unabl captur type modi cation relat special construct sampl question presid russia presid largest countri electron biggest export japan unit phrase correct wrong modi cation relat countri lake countri keyword phrase point correct answer construct mean critic depend relat di er mean queri involv common word salient keyword commonli occur keyword index return larg number result salient keyword question word word stopword discard standard keyword search engin return sentenc keyword sentenc answer user question chapter futur work appli natur languag process techniqu show potenti improv qualiti current inform retriev system current system prototyp hope serv core devot integr natur languag process techniqu inform retriev outlin roadmap futur direct build gener inform access system inform extract question answer system todai util inform extract techniqu import compon architectur question inform extract technolog match semant type expect user queri semant type answer analyz anticip type answer question answer question locat question answer system focu retriev text fragment correct answer type techniqu sentenc multipl construct dicult distinguish sentenc supposedli answer question di erent dicult determin semant type question standard question answer varieti di erent construct inform extract techniqu determin valid integr technolog inform access system improv inform access system util inform extract technolog preprocess corpu text order gener accur relat natur languag full entiti person handl linguist parser lead incorrect analysi unabl henri lead construct incorrect tree inform extract technolog appli ecient preprocess construct recogn extract integr inform extract system relat index accuraci relat deriv inform extract techniqu enabl inform access system intellig handl variat entiti john person form relat involv match relat altern titl person correspond establish locat compani busi date textual marker techniqu fail succe answer question involv easili identi textual marker answer sky scatter ga molecul dust particl air give sky blue system care equat di erent context account wrong equat occurr rst entiti variat entiti handl combin heurist rule rule determin dictionari lookup list variant sophist queri process system bene analysi user queri order identifi relat expect correct answer incorrect answer question ant retriev armi ant live hunt insect fact live ant vari greatli wai life worker ant live month die match algorithm convert question wildcard chapter perform addit process base type sentenc return correct answer ant live relat sophist queri process map preposit phrase head preposit allevi techniqu distinguish sophist match algorithm drawback standard boolean model inform retriev incorrect assumpt keyword equal import lead develop inform retriev model util model fuzzi logic exibl achiev perform approach util exibl inform access model base relat di erent type equal ect match strategi score system relat gener import modi cation illustr question black bear system present retriev knowledg habitat type bear random fact black bear big modi cation relat frequent strict match produc fewer prevent inform overload match algorithm rel import relat account lead system sati es user inform resolut disadvantag perform retriev sentenc level retriev techniqu tradit inform retriev inabl handl anaphor refer retriev sentenc level good basi question answer sentenc serv unit inform easili digest user answer question span problem allevi provid result surround sentenc success inform access system handl anaphora explicitli equat refer express entiti refer substanti bodi research anaphora rang tradit approach recent approach resolut appli question answer system integr approach resolv express de nite noun phrase enhanc perform question answer system express initi introduct subsequ refer entiti accomplish einstein physicist left germani unit state ibm announc plan pronoun refer express common type handl sophist question answer system de nite noun phrase de nite noun phrase refer entiti introduc head noun refer express complet bush presid instanc easi explicit link exist refer express entiti refer de nite noun phrase anaphor dicult bird capit taiwan citi resolv de nite noun phrase previou exampl requir knowledg capit citi larg semant ontolog entiti requir properli resolv de nite noun phrase anaphor exist resourc util solv problem wordnet primarili lexic semant classi cation scheme semant ontolog match classi cation scheme implicit ontolog ontolog resolv de nite noun phrase refer built requir time construct potenti techniqu simplifi ontolog build process extract ontolog de nition corpu corpu de nition color hypernym pair extract automat limit relationship entiti obviou human explicitli american person signi canc refer express plai import role resolut modul import compon complet question answer system relat queri techniqu classic amelior diculti synonym text retriev system research shown queri involv synonymi relat gener ectiv accur disambigu relat index distinguish word identi cation syntact relat appli relat question syntax er clue sentenc chapter bank river bank river transform rule system implement modul transform aspect system requir attent transform rule solut problem linguist provid gener set broad coverag rule construct major challeng creation rule transform rule impract build larg knowledg base rule sole hand techniqu includ util distribut system gather linguist applic statist machin learn techniqu hybrid system success question answer system hybrid combin varieti rang sophist languag analysi robust techniqu model index potenti recal inform access keyword statist techniqu appli boost recal accept level inform retriev scheme base keyword statist robust integr question answer system util mechan method result deriv method complet inform access system requir integr di erent techniqu synergist ect hybrid bring step closer inform access system tomorrow chapter document describ architectur integr natur languag process techniqu inform retriev ternari express fundament structur strike balanc sophist natur languag techniqu ecienc index retain advantag natur languag expand domain unrestrict text initi implement show dramat improv exist system limit test architectur serv good foundat natur languag question answer system bibliographi dougla appelt david israel introduct inform extract technolog avi van der cha van inform retriev inform process decemb avi van der cha van evalu index scheme colloquium ir probabl applic inform retriev annual review inform scienc eric john lisa david marc mani sy call text retriev confer william cooper bool inform process bruce croft david lewi approach natur languag process document retriev annual intern acm sigir confer research develop inform retriev dagan automat process larg corpora resolut anaphora refer annual intern confer comput linguist joel fagan experi automat phrase index document comparison syntact method phd cornel robert kevin humphrei combin approach question answer larg text collect riao confer multimedia inform access georg natur languag input simul program system technic report naval postgradu hobb pronoun resolut technic report depart comput citi citi karen georg stephen editor natur languag approach kluwer academ bori katz english index retriev ph winston sa arti cial intellig expand volum mit bori katz annot world wide web natur languag riao confer comput assist inform search internet bori katz beth levin exploit lexic regular design natur languag system intern confer comput linguist bori katz patrick winston gener english commut transform technic report mit arti cial intellig bori katz patrick winston natur languag interfac european confer integr interact comput system kennedi anaphora anaphora resolut parser annual intern confer comput linguist kupiec robust linguist approach question answer encyclopedia annual intern acm sigir confer research develop inform beth levin english verb class preliminari investig chicago lin parser intern confer comput linguist lin annual meet associ comput linguist lin minimalist parser maryland linguist colleg march kenneth semant relat tripl text retriev confer edward appli semant relat extract inform retriev master massachusett institut georg miller lexic english commun thoma morton question answer text retriev confer john eric predict annot question answer text retriev confer cs sidner comput theori de nite anaphora english technic report mit arti cial intellig alan index structur deriv syntax system descript text retriev confer wei li inform extract support question answer text retriev confer jim fang jose troi jin jon wild natur languag inform report text retriev confer strzalkowski peter evalu preliminari report recent advanc page kluwer academ cj van rijsbergen theoret framework inform retriev acm confer research develop inform cj van rijsbergen logic inform retriev comput jose antonio anaphora resolut question answer system annual meet associ comput linguist ellen wordnet disambigu word text retriev annual intern acm sigir confer research develop inform retriev ellen queri relat annual intern acm sigir confer research develop inform retriev ellen dawn tice question answer track evalu text retriev confer bl webber formal approach anaphora patrick winston arti cial intellig david evan evalu syntact phrase index nlp track report text retriev confer 
IR/1295 5 approach web inform extract nichola kushmerick comput scienc colleg introduct inform agent distribut system receiv goal user gather inform relev goal varieti process content deliv result user focu stage gener architectur survei varieti inform extract techniqu enabl inform agent automat gather inform heterogen sourc agent mediat request satisfi agent retriev address inform geograph servic freight forward serv request quot relev freight retriev duti legal constraint govern weather inform estim transport inform extract form shallow document process involv popul valu automat extract document past research develop rich famili gener techniqu suitabl wide varieti rigidli format document html gener automat document newspap articl email messag view inform extract core enabl technolog varieti inform agent focu specif inform tangenti agent discov relev sourc verifi authent retriev cach polici minim commun ensur fresh observ xml elimin automat inform extract terabyt content numer legaci servic export data xml imposs determin correct annot applic idiosyncrat unit currenc includ extract split date reason expect automat inform extract continu essenti year scalabl kei challeng automat inform extract relev dimens dimens abil rapidli process larg document collect system gener scale regard reli simpl shallow extract sophist natur languag process problemat dimens number distinct sourc agent request quot thousand freight weather inform dozen forecast challeng scenario sourc format content sourc requir custom set extract rule machin learn approach scale dimens chapter machin learn enabl adapt inform extract system automat learn extract rule train data order scale number sourc gener idea adapt inform extract human expert annot small corpu train document fragment learn system gener exampl produc form knowledg rule reliabl extract content document train data assumpt underli adapt easier annot document write extract requir program describ techniqu aim minim amount train data requir elimin manual annot adapt inform extract research commun develop wide varieti techniqu tailor extract task document type approach web inform extract categor learn extract knowledg structur formal regular grammar automata survei promin addit research relat entir wrapper core learn section introduc wrapper form adapt tailor highli regular web section describ varieti express extens basic wrapper induct section explor techniqu natur languag section describ hidden markov model adapt section discuss approach wrapper verif central issu adapt section describ techniqu extract fragment coher section explain activ learn techniqu minim amount train requir accur gener sake breviti describ techniqu inform idea wrapper induct kushmerick formal adapt web inform extract work wrapper induct kushmerick identifi famili wrapper demonstr wrapper express learn wrapper numer web effici hand train cpu illustr wrapper induct web page shown figur html encod content extract extrem exhibit featur salient discuss wrapper consist sequenc delimit string find desir content simplest case figur content arrang tabular format wrapper scan pair delimit total delimit notat delimit delimit case wrapper execut procedur lr scan string begin scan ahead occurr procedur extract text posit column row procedur scan extract text posit column row process start extract termin miss end figur formal idea wrapper class lr wrapper consist set pair column lr provid procedur procedur scan begin scan ahead occurr procedur extract text posit column row lr scan extract text posit column row process repeat column search procedur start extract termin miss end lr machin learn task automat construct lr set train document lr learn delimit learn independ kei insight candid valid delimit impact delimit base kushmerick describ algorithm learn lr wrapper algorithm simpli potenti valu select satisfi constraint wrapper work correctli train data kushmerick demonstr theoret pac algorithm requir modest train sampl converg correct wrapper effici learn algorithm exist wrapper discuss limit lr class show handl document complic format simpl lr class successfulli wrap web survei countri procedur occurr scan occurr save posit start attribut scan occurr save posit end attribut return extract pair procedur page occurr scan occurr save posit scan occurr save posit return label fig fictiti internet site provid inform countri telephon countri web html document content lr gener gener lr lr mean definit solut web inform demonstr simpl techniqu remark effect lr effect simpl minor complic format render lr ineffect lr class requir reliabl begin attribut delimit figur modifi includ head code top case delimit lr work fact show legal lr wrapper document modifi manner kushmerick tackl issu extend lr famili addit wrapper class class addit delimit skip text head tail page head delimit skip top enabl work correctli class addit delimit identifi entir tupl regular lr strategi extract attribut turn idea combin fourth wrapper class kushmerick explor simpl wrapper data format simpl tabular fashion class extract tabl content oper lr process possibl level continu level return level return level attribut class combin kushmerick develop special learn algorithm class empir complex express power wrapper class extent effici learn class successfulli wrap survei algorithm learn wrapper time grow exponenti number pac analysi reveal requir substanti train exampl converg compar class express wrapper class initi investig lr famili substanti research effort elabor altern wrapper deriv effici learn algorithm extend wrapper class numer limit muslea al hsu dung develop algorithm address miss attribut complic page involv miss null attribut valu delimit simpl wrapper process remaind page correctli french site countri address franc attribut simpl wrapper class discuss assum simpl relat model attribut singl structur attribut natur scenario hotel guid explicitli list citi serv wast binari encod citi multipl attribut order wrapper assum attribut occur fix variant order abound complic document movi site list date titl movi prior titl recent movi disjunct delimit wrapper discuss assum singl delimit complic site multipl delimit site list price bold sale price render red nonexist delimit wrapper earlier assum irrelev background token separ content assumpt violat case depart code separ number string problem relev asian languag word token space typograph error except document error occur format drive simplist wrapper fail entir page small portion badli format sequenti delimit wrapper class assum singl delimit simplest develop accur wrapper scan delimit sequenc extract restaur review simpler scan scan final scan forc wrapper scan document singl delimit reliabl extract content hierarch organ data nest class step handl result larg neg complic scenario extract nest structur hsu dung address problem learn wrapper express class determinist transduc formal handl requir mention transduc process document extract singl extract control return start state tupl extract attribut repres pair state identifi start attribut identifi end gener automaton model state connect arbitrari permit miss attribut attribut multipl attribut order path start end govern express rule languag disjunct delimit limit form system recov format error except hsu dung describ algorithm effici learn wrapper transduc train data report wrapper class handl site wrap wrapper class muslea al identifi class wrapper unlik hsu tackl issu mention main distinguish featur muslea wrapper multipl delimit call landmark insist exist singl delimit identifi relev posit deep insid wrapper sequenc delimit jump posit seri simpl step simpl step easier enabl robust extract major featur muslea work formal nest organ data express simpl hierarch approach kushmerick extract natur text techniqu aim highli regular html emit cgi program research inform extract natur email newspap result relev structur domain recent investig shown result freitag kushmerick explor wrapper defin class extract pattern essenti lr case attribut enrich class permit delimit token type specif instanc corpu email seminar algorithm learn rule extract start match document pm fragment extract underlin rule basic find start find end time token final simpl rule languag extract free text freitag kushmerick improv perform boost gener techniqu improv accuraci weak learn learn rule individu rule high low rule set high high recal result accur extract algorithm competit approach varieti superior boost wrapper induct perform essenti perfectli task extract seminar announc competitor attribut speaker seminar locat soderland describ relat approach techniqu inform extract free text extract rule correspond restrict class regular express regular express serv contextu pattern determin fragment delimit determin boundari target ment languag import design work document span spectrum unstructur natur text highli structur web page depend structur train learn algorithm automat creat pattern simpl extract suffici accur learn algorithm bother add addit contextu constraint extract price number bedroom list document br incl upper system learn rule portion regular express valu extract rule extract content document hidden markov model work freitag kushmerick soderland instanc gener approach rigidli structur html document structur document email newspap articl approach brittl facil evalu strength evid guid extract held seminar document typograph error held techniqu make binari uncertain evid hidden markov model principl effici approach handl sort inher uncertainti hidden markov model stochast automaton state emit token fix transit state occur fix distribut hmm attract comput devic effici algorithm learn distribut infer state sequenc observ token sequenc hmm inform state token extract email seminar announc hmm state start time end time speaker locat token addit state gener token perform standard hmm viterbi decod algorithm determin gener observ extract fragment simpli read path hidden markov model successfulli numer research varieti extract scenario kei challeng effici algorithm determin state topolog distribut probabl forc permit initi work gener state connect manual evalu train corpu fig snapshot altavista search site note present content chang attempt automat learn topolog gener approach greedili search space topolog maxim object function seymor al attempt maxim probabl train data topolog approach effici potenti goal hmm model train data perform accur extract freitag mccallum object function actual accuraci topolog extract valid corpu approach significantli slower result compact topolog gener wrapper mainten work earlier ignor import complic inform agent gener control sourc receiv data wrapper tend invari reli idiosyncrat format detail observ learn process sourc modifi format user observ regular longer hold wrapper fail concret figur show altavista search site redesign kei challeng wrapper mainten wrapper verif wrapper oper wrapper challeng wrapper verif difficulti web content format verif algorithm distinguish chang microsoft stock price check time extract valu intuit verif algorithm realiz valu outlier defect wrapper kushmerick describ simpl accur algorithm wrapper verif algorithm learn probabilist model data extract wrapper train period oper correctli model captur properti train data length fraction numer charact extract data verifi wrapper train extract data evalu learn model estim probabl wrapper oper correctli algorithm domain independ ti wrapper class learn treat wrapper inspect output algorithm handl xml relat applic wrapper class wrapper receiv attent al learn probabilist model extract data similar substanti express kushmerick sensit model enabl wrapper wrapper deem learn model identifi probabl target fragment document train data remov data wrapper induct algorithm al demonstr empir approach highli accur extract scenario extract content work highli simplifi task assum involv simpli process document extract target fragment extract inform extract distribut multipl attribut page relev extract object figur show simpl scenario attribut valu multipl extract valu harvest collect hyperlink document issu handl wrapper class defin earlier muslea catalog formal permit extract fragment multipl object inform extract commun long investig issu refer approach requir linguist process applic shown figur jensen cohen address problem languag extract data rule express raw extract data group larger composit object jensen cohen argu languag suffici express handl data extract web site export job product suggest algorithm automat learn rule exampl group data kei bottleneck adapt inform extract obtain label train data machin learn motiv fact cost label fig complic extract task attribut valu distribut multipl object document cost write extract rule hand label document requir domain gener tediou approach simpli assum adequ train corpu research effort investig method minim amount train data requir achiev satisfactori level gener basic idea activ learn start small amount train run learn learn wrapper predict remain unlabel document learn system gener addit train document trivial corpu duplic learner suggest document annot activ learn context wrapper muslea al basic idea approach inform extract task correl origin task dual system identifi unlabel document recal muslea wrapper learn algorithm learn sequenc landmark scan begin document start fragment extract altern find posit scan backward end document set landmark extens solv learn task parallel train data result wrapper appli unlabel document system user label document wrapper give answer wrapper unlabel document subsequ learn muslea al demonstr version algorithm requir significantli train data obtain level gener averag varieti challeng extract error learn wrapper user annot ten train document chosen intellig compar ten document chosen randomli brin explor sort extract user system exampl concept learn extract book user suppli small sampl robot job extract system flesh list addit instanc algorithm iter search web seed pair find document learn inform extract pattern appli pattern remaind page result extract pair ad seed process iter process converg extract pair correct preliminari experi demonstr result al focu bigger wrapper learn train page onlin bookstor return page format differ content extract intuit approach wrapper learn compar page find similar differ similar correspond common format structur differ correspond data extract repeatedli replac differ note repetit algorithm learn wrapper correspond regular manual label train data al report algorithm work varieti domain summari inform extract core enabl technolog wide varieti agent central challeng inform extract abil scale number varieti inform sourc varieti adapt inform extract approach machin learn techniqu automat learn extract rule knowledg train data due highli practic natur approach chapter test real world exampl mean gap pure research practic usag agent system smaller glanc adapt inform finit state techniqu learn extract knowledg regular mar automata addit core adapt briefli discuss issu relat entir inform extract fundament open issu adapt inform extract method determin techniqu suit extract task complic judgment requir methodolog heterogen document measur order predict simplest approach deliv satisfactori perform acknowledg bernd thoma discuss refer weischedel learn proc conf appli natur languag brin extract pattern relat world wide web proc sigmod workshop merialdo automat data extract larg web site vldb page freitag kushmerick boost wrapper induct seventh nation confer page juli august texa freitag mccallum inform extract hmm structur learn stochast optim seventh nation confer juli august texa hsu dung gener transduc semistructur data extract web inform jensen cohen group extract field proc workshop adapt text extract kushmerick wrapper induct inform extract phd kushmerick regress test wrapper mainten proc nation confer artifici page kushmerick wrapper effici express artifici kushmerick wrapper verif world wide web doorenbo wrapper induct inform extract fifteenth intern joint confer artifici volum page august japan leek inform extract hidden markov model san minton learn common structur data proc nation confer artifici muslea extract pattern inform extract survei proc workshop machin learn inform knoblock hierarch approach wrapper induct proc intern confer autonom page knoblock select sampl redund view proc nation confer artifici rosenfeld learn hidden markov model structur inform extract proc workshop machin learn inform soderland learn inform extract rule free text machin 
Agents/828 1 agent vickrei auction consid special track theori felix brandt gerhard wei institut fur technisch universitat munchen germani tel abstract recent year auction import field multiag system mechan resourc alloc task assign case vickrei auction protocol prescrib individu agent interact order agreement show vickrei theoret inappropri agent particip auction process attitud econom agent make reduc profit competitor main goal maxim profit introduc agent deviat domin strategi paper present strategi bidder repeat vickrei auction intend loss fellow agent order absolut group bidder strategi evalu simpl task alloc scenario introduct area multiag system concern system technic entiti call agent interact intellig achiev steadili grow interest past decad kei problem address area autom resourc alloc task assign individu agent solut problem common practic appli result insight auction theori understood auction protocol english dutch vickrei auction vickrei auction receiv attent multiag commun appli varieti context oper comput network vickrei auction favor main requir low bandwidth time consumpt possess domin bid true valuat section auction privat remain characterist make vickrei auction protocol appeal point view autom vickrei origin formul sell good resourc work bidder make seal bid express amount bidder submit highest bid win price pai winner equal highest bid task assign scenario vickrei auction work round reason refer vickrei bidder execut task make bid express amount pai task bidder submit lowest bid win winner receiv amount equal lowest bid payoff lowest bid minu prime cost win winner pick randomli paper concentr vickrei auction task assign present result hold vickrei auction origin formul import bid seal bidder win auction necessarili lowest bid low bid gener assumpt made applic multiag system individu agent intend maxim profit care profit mean case set observ maxim delib loss rival compani minim profit practic compani accept lower profit sell good loss damag compet compani bind gain costum compani exhibit behavior attitud consid period time absolut profit import profit rel compani profit worth point behav ration act condit object maxim absolut paper investig perform vickrei auction multiag system presenc agent exclud relat problem discuss econom literatur bidder extern describ prefer bidder win auction sell patent nuclear weapon approach sole agent care bui good award task contract interest fellow profit strengthen market posit major differ paper structur section explain agent aim maxim absolut profit bid true valuat vickrei auction section formal captur attitud sketch section show impact bid strategi section introduc analyz strategi repeat vickrei auction section present result illustr implic strategi section conclud paper overview advantag disadvantag agent domin strategi vickrei auction domin mean agent appli strategi receiv highest matter strategi bidder domin strategi bid true valuat task agent bid bid privat valuat condit exist domin strategi equilibrium bidder symmetr independ valuat task impli task agent privat valu submit bid profit agent defin equat profit easili bid task optim strategi agent investig profit make bid privat suffic model agent repres entir competit care win lose draw distinct fellow bidder bid prime cost condit agent bs bid win auction receiv monei bid lose win bid win gain task price remain payoff bid case describ result paid amount monei bid monei prime lose lose reduc bs payoff bid yield profit bid true valuat extrem simplifi bid due absenc wast dutch auction attitud multiag applic assum object agent team maxim absolut profit care profit made bidder dont estim privat domin equilibrium exist independ risk neutral agent applic realist assum agent present gain monei rel scenario wise avail agent accept small loss great loss agent make develop formal descript attitud start point reason assum agent maxim differ profit gain mean profit hand agent loss hand consid equal import point view agent view captur agent intent rival achiev higher flexibl describ analyz agent object harm competitor agent put emphasi profit loss lead formal specif agent agent agent maxim weight differ profit profit competitor agent intend maxim payoff payoff profit paramet call rate rate crucial formal agent behavior obviou formula cover agent set higher hurt greater prioriti pure destruct agent defin agent balanc profit profit competitor equal import vickrei auction implic formal notion attitud vickrei auction enorm remain implic theoret investig combin equat payoff agent function bid payoff case section agent effect win price agent receiv complet depend bid carefulli adjust payoff agent simpli profit bid capabl reduc bs profit bs privat optim strategi bid figur reduc bs profit absolut minimum fig reduc bs profit minimum prime cost prefer great loss gain small profit rate bid reject gain make lose rate bid safe figur top monei lose fig aggress bid bs strategi bid bid strategi agent unrealist assumpt agent privat valu lowest privat omit margin includ avoid random result occur bidder share minimum bid strict version theorem state vickrei auction balanc bidder strategi defin equat nash equilibrium assumpt state agent appli reason singl agent deviat payoff agent lowest privat lowest bidder repres lowest bidder suffic differenti individu bidder vickrei auction sole victor possibl agent harm cheapest provid hurt highest bidder inferior agent reduc profit lowest bidder assum appli strategi defin payoff payoff payoff payoff max ba payoff bid receiv equal bid payoff diminish strategi defin equat equilibrium independ rate bidder claim strategi optim strategi reduc loss occur worst case encount mean bidder singl agent reduc agent payoff min payoff min yield minimum profit win yield minimum profit lose auction maximum minima max ba min payoff max ba min payoff min payoff due fact equilibrium point comput set bid repeat auction basi theoret foundat previou develop bid strategi realist environ gener agent privat bidder principl agent possibl figur instanc mean care auction learn previou auction paper deal techniqu reveal privat valu auction fix number repeat round balanc agent lose auction round task auction bid win auction receiv amount equal lowest privat cheapest agent agent appli domin figur privat place bid middl privat valu lose differ valu safe cut competitor profit auction round total number round high invest pai scenario agent follow domin bid strategi bid strategi bid bid bid eleg agent appli strategi agent win paid monei lowest bid produc huge deficit step step approach possibl reduc bid round round small margin agent bid reach figur displai modifi strategi step size equal privat algorithm emul aggress strategi algorithm work stabl dynam environ agent vanish time time strategi balanc agent appli agent reduc win agent profit figur real privat cheaper agent time agent bid won lost start bid bid lost lost won won lost won bid bid fig strategi repeat vickrei auction group agent bidder risk rate prescrib care agent small step size order safe competitor suffer huge loss make neg profit reason set depend number distribut privat valu rate upper bound level commit contract task execut contract bind breach pai penalti commit contract unavoid loss agent produc cheapest competitor reduc break neg contract due fact reason close deal figur privat agent incent accomplish task break contract loss make accept contract greater penalti pai break deal common definit penalti fraction contract agent breach contract pr actual task price pr penalti rate give assumpt pr agent break contract task price equal privat distribut prime cost true case result set investig paper similar due reason limit present experi conduct detail result includ random cost tabl number buyer execut task associ prime cost task execut interest task price higher cost price bid integ valu sell task interest calcul submit seal bid submit lowest bid declar winner lowest bid price announc paid price execut task equal win winner pick randomli earn monei handl privat task prime cost assum execut task round commit tabl prime cost ident abil handl cheapest price bid privat valu gain horizont line figur show profit accumul task task task tabl fair cost tabl round ce ce appli domin strategi bid prime cost ce harm competitor reduc profit minimum ce agent rate larg step chose care step size set reason ce set visual strategi work contrast normal case appli domin strategi make equal ce outperform rival lose sum profit entir group reduc singl agent aggress strategi vulner vickrei auction bid glanc agent care profit make highest profit odd effect explain fellow bidder auction round fig ident ad equat ce risk entir profit order hurt ce ce complet gain perform break expect figur agent cut quarter profit rival due mutual rate attitud agent formal introduc paper lead import strateg behavior agent argu impli exist agent paper vickrei auction show auction addit defici vulner bidder effect polici enabl easi price manipul common english auction privat bidder vickrei auction strategi paper work english auction inabl prevent profit reduct regard major disadvantag auction type dutch auction suffer strategi auction round fig ident problem strategi repeat vickrei auction desir intend cut cheapest agent profit highest margin reduc bid bidder stai bid privat lose monei harm round requir agent make special feasibl set behavior paper opposit bidder bidder coordin bid order agent bid intent harm fellow bidder research step explor strategi agent detail plan examin strategi complex environ number vari time task intend investig type collabor agent act effici refer brandt bid repeat vickrei auction technic report institut fur technisch universitat wei task assign multiag system base auction level commit contract klusch cooper inform agent volum lectur note artifici artifici page berlin weiss user control mode ip alloc present mit workshop internet econom ke ms miller incent engin comput resourc manag ba ecolog comput ferber system introduct distribut artifici intellig john wilei son huberman sh clearwat multiag system control build environ intern confer multiag system page menlo aaai press sell nuclear weapon american econom jk hr varian price internet public access internet confer school rp mcmillan auction bid journal econom pr rj weber theori auction competit bid ohar nr editor foundat distribut artifici intellig john wilei son game inform basil mh rothkopf rm model cheat vickrei auction journal mh tj ep kahn vickrei auction journal polit tw sandholm limit vickrei auction comput multiag system intern confer multiag system menlo aaai press tw sandholm vr lesser issu autom negoti electron extend contract net framework intern confer system page tw sandholm vr lesser advantag level commit contract protocol nation confer artifici intellig page sun microsystem manag share resourc market mechan vickrei counter competit seal tender journal ca jo ws distribut comput economi ieee transact softwar editor multiag system modern approach distribut artifici intellig mit 
IR/1622 5 build effici wireless sensor network £ john heidemann ý ramesh govindan ý ý scienc institut admir marina del usa distribut node commun leverag topolog locat node independ applic investig emerg class distribut system commun reli network topolog locat commun base attribut extern network topolog relev applic combin kind data enabl process data collabor signal similar problem approach essenti emerg applic sensor network resourc bandwidth energi limit paper descript softwar architectur support data process show approach aggreg nest queri significantli affect network traffic experi aggreg reduc traffic nest queri reduc loss rate aggreg studi paper demonstr nest queri form present evalu approach oper testb introduct distribut node £ work support darpa grant part nsf grant part made part due support cisco system fabio silva ý ý deborah estrin þ comput scienc depart lo angel lo usa leverag topolog locat node independ applic commun built commun primit level bind servic map topolog requir internet ip address provid suitabl rout ip address assign address node topolog proxim drawn address prefix logic connect distinct physic topolog assign essenti scale rout system carri forward dn hierarch node system implement ip web search engin provid document object content distribut network add geograph constraint system jini layer approach resourc discoveri ip network devic investig emerg class distribut system commun reli network topolog locat commun base extern network topolog relev base capabl sensor type geograph locat approach kind effici elimin overhead commun requir resolv bind data enabl activ process insid data reduct data gener benefit appli internet bandwidth delai throughput process primari constraint technolog trend condit wireless sensor network sensor network predic assumpt feasibl small devic memori process wireless addit sensor sensor network process time bit commun plenti fast bandwidth bandwidth scenario kaiser observ instruct execut energi cost send bit radio environ encourag comput reduc commun fewer level indirect messag process opaqu packet essenti design sensor network resourc constraint wireless sensor network met system extern frame refer tradit approach build underli topolog scheme ip multipl layer bottleneck ten overhead unreason hundr thousand node vari avail movement domain sensor network profit elimin multipl layer rout data directli term effici attribut base extern frame refer attribut geographi sensor type reduc level bind rout reduc resourc consumpt addit process essenti sensor network suggest comput collabor signal process occur close data collect process depend data trigger attribut data type process data kind process similar activ network differ oper environ sensor network solut illustr process sensor wireless monitor system mixtur light motion sensor vigil sensor microphon camera conserv energi bandwidth audio sensor trigger light sensor comput distribut network queri label sensor type system design time queri network handl node match sensor relev geograph region applic hear relev sensor respond sensor trigger move network handl directli light audio sensor altern architectur central directori activ sensor central applic interrog monitor specif trigger goal elimin commun cost maintain central inform provid robust network spite chang move limit batteri power explor approach work section quantifi potenti save section demonstr exist simpl architectur commun achiev highli energi effici applic design kei contribut work identifi specif scheme flexibl match rule ground share framework attribut sensor type show approach enabl process local data quantifi benefit run system previou work discuss commun primit constitut direct work understand design space network protocol underli direct evalu perform find scalabl good number node traffic work develop softwar architectur realiz attribut process oper system emploi simplifi attribut scheme aggreg simul necessit approxim effect radio paramet simul set match sensor network hardwar paper evalu design question process encount deploi sensor present result data testb detail implement mac numer earli system develop gener approach softwar design sensor network work uniqu replac augment underli network rout match rule effici implement express cover wide rang process relat work work build prior work sensor network system larg amount work gener special domain intern structur mechan servic research industri develop numer system layer top network univ arizona design provid servic discoveri group comput print unload includ attribut build standard internet protocol commun commerci system ldap oper internet rout provid primarili hierarch organ depend address rout limit add substanti overhead appli system highli environ sensor network approach servic locat smart space requir servic ip host servic registr process system provid process altern provid system commun structur distribut system linda structur distribut program cpu common memori call tupl space implement basic commun implement assum uniform rapid commun processor system isi inform bu provid approach inform provid publish inform client subscrib subset inform system design robust assum commun node approach directli applic sensor network process process close process manual configur specif work primit solut specif problem srm suggest data fundament data unit reliabl multicast demonstr approach distribut whiteboard work inspir differ provid wider rang match oper ad process leverag sensor oper directli commun protocol internet multicast infrastructur process recent work activ network activ servic examin wai provid process internet sampl applic includ inform network cach work built augment extend assum node individu address build directli commun primit identifi data node work differ activ servic assum commun cost node vari greatli activ servic assum roughli distanc node differ activ network primarili target target sensor network bandwidth energi comput power compar plenti activ network typic consid domain bandwidth ratio comput power bandwidth energi issu approach distribut code question code safeti portabl problem central sensor network devot singl complex network benefit execut environ support upgrad recent work adapt web cach file share system explor process unlik activ network approach protocol design applic work run directli commun virtual network layer internet system sensor network research activ advanc sensor node radio hardwar work instrument clarifi comput commun process focu process motiv work work base sensor primari differ work structur data commun internet ad hoc rout al survei protocol sensor network ad hoc rout recreat requir kind directori servic locat unlik approach attribut ad hoc rout support process jini resourc discoveri system built internet protocol directori servic java distribut process user make suit network high bandwidth multicast distribut directori network process intermedi system address problem wireless network servic discoveri servic locat object network collabor server target high bandwidth resourc work present fundament advanc network commun network devic work static hierarchi network host similar tier architectur full data dynam process spin evalu variant flood wireless sensor network data spin identifi metadata assum individu sensor address attribut data global uniqu identifi spin process intent system system oper overlai network internet attribut structur mechan method cope dynam locat devic similar approach motiv mechan primari differ assum commun basic commun primit construct overlai network internet architectur impli distribut match small commun node manag resolv cooper manag part namespac detail match system work sophist hierarch attribut ing procedur approach modest comparison smaller add compar oper addit equal analyz perform rout mechan data compress commun compress reduc energi consumpt compress kind process support flow opportun aggreg work mechan activ dataspac describ attribut base mechan queri physic object produc store local data dataspac divid smaller administr logic logic group address network level multicast address correspond geograph valu attribut serv network indic queri result involv aggreg specif queri address approach similar map attribut geometri larg number multicast group rout directli attribut indirect explor process devic system distribut queri sensor network move data central site sensor data repres abstract data type public interfac correspond specif signal process function support sensor type perform join aggreg network central comput queri plan work common emphasi studi nest queri inspir work primari differ work placement process determin filter nest queri enabl placement central translat queri assign process distribut incur overhead central collect network inform queri optim declar rout lincoln lab closest work api defin collabor develop independ implement primari differ work focu process evalu work complet section architectur commun architectur base direct match filter direct dissemin inform distribut system data manag list tupl match rule identifi data arriv intermedi filter process data approach provid extern framework relev applic compon balanc gener servic match rule attribut filter describ compon direct direct data commun mechan sensor network data sourc sink attribut identifi inform provid interest goal direct establish effici commun sourc sink direct commun paradigm commun tradit network describ element simpl sensor network design track anim wilder user network track movement anim remot park direct track task repres interest interest list pair describ task scheme describ detail attribut attribut describ data desir sensor type possibl geograph region identifi contact relev sensor term sink denot node origin interest destin data interest propag sensor node region kei featur direct sensor node node store interpret simpli forward sensor node receiv interest rememb neighbor neighbor interest set gradient gradient repres direct data match interest statu demand activ inact possibl desir updat set sensor node redistribut interest neighbor node infer potenti sourc geograph inform exist similar interest forward subset neighbor simpli broadcast interest neighbor sensor node match interest applic activ local sensor begin collect data activ expect sensor sensor node gener data messag match interest direct data repres scheme sensor node gener event descript term sourc data cach intermedi node propag sink cach data level core mechan cach suppress duplic messag prevent preferenti forward interest core primarili interest exact hash attribut comput compar complet cach data process data detect singl object sensor merg singl base criteria initi data messag sourc mark exploratori neighbor match gradient sink multipl receiv subsequ data messag interest prefer neighbor deliv copi data sink reinforc prefer turn reinforc prefer upstream event sourc interest interest propag sink event sourc gradient initi gradient set node prefer path sensor node attempt local repair fail path sink neg reinforc current prefer neighbor neighbor deliv sensor data neg reinforc propag remov gradient exist path longer neg reinforc suppress loop duplic path due network dynam initi exploratori data subsequ messag reinforc path period sourc send addit exploratori data messag adjust gradient case network node energi temporari network recov lost exploratori messag recoveri data loss left applic simpl applic transient data sensor report state addit recoveri develop retransmiss scheme applic transfer persist data object simplifi descript point kei featur differ tradit network commun sensor network interest data commun unlik tradit data network commun node sensor network corollari previou observ sensor network sensor node interpret data interest messag design choic justifi sensor network sensor network commun network node global uniqu identifi global uniqu address regular oper distinguish neighbor individu node process perform coordin close phenomena perform data result energi save describ usag direct paradigm type figur paradigm gener discuss applic section attribut tupl match rule messag applic interest tupl attribut identifi uniqu sink event sourc figur simplifi schemat direct data deliveri reinforc path attribut set attribut formal match attribut actual compar match true match return return true sink figur match algorithm kei drawn central author practic implement simpl number assum coordin internet protocol number attribut implicitli data format float point valu uninterpret binari data oper field defin data messag interest interact oper usual binari comparison match user actual oper formal comparison paramet comparison match compar formal paramet attribut set actual formal paramet miss match actual attribut set match fail gt actual match lt gt set attribut complet match match succe direct attribut set match match algorithm succe match style similar rule languag linda add match rang oper addit equal multipl attribut oper present effect formal satisfi match success approach strike balanc implement flexibl simpl bound set oper implement ten line code rectangular region attribut match continu section user sensor network track anim queri translat interest attribut eq interv durat ge le ge le implicit attribut ad identifi messag interest interest specifi detect anim region rectangl inform frequent data return long queri sensor network program anim search routin time download mobil sensor watch interest anim express interest interest attribut eq type interest arriv sensor activ sensor paramet provid repli detect sensor detect data messag includ attribut instanc intens confid timestamp class messag satisfi origin interest encod attribut addit inform confid sender detect illustr detail specif queri show data conveni encod geometri attribut simpl match rule work applic applic subset omit geograph constraint small sensor singl attribut sensor found primit provid good build block rang describ section match perfectli cover scenario task simpl match case approxim code refin choic perfect rectangl align coordin system insuffici describ arbitrari geometr shape shape accomplish multipl smallest bound rectangl applic ignor request insid rectangl requir region applic gener attribut clarifi paramet subtyp filter applic influenc process filter filter mechan code run network assist process applic provid filter sensor principl filter distribut mobil code packag filter regist kind data handl trigger time kind data enter node filter arbitrarili manipul cach influenc gener messag filter access intern inform includ gradient list neighbor node filter typic tive signal similar task benefit control data movement addit found debug monitor continu filter suppress concurr detect anim sensor regist interest detect interest data attribut record desir interv repli interv unit suppress repli sensor sophist filter count number detect sensor add addit gener kind aggreg rate applicationspecif manner filter discard reduc unnecessari commun greatli extend oper lifetim describ applic filter section quantifi benefit aggreg scenario section implement implement part architectur current refer implement version compon lab implement attribut match filter implement run linux desktop pc sensor node cpu ram flash figur sensor node node custom figur implement bare subset servic design run tini processor memori sourc code implement found web implement build simpl radio api support broadcast unicast neighbor neighbor kind requir persist persist identifi ethernet mac oper assign identifi basic api refer implement includ network rout api summar figur complet specif sourc api defin approach data handl receiv node subscrib set attribut subscript result interest network set gradient callback function invok relev data arriv node applic gener inform publish send specif data attribut publish call match subscript activ publish data leav node optim sensor node avoid gener data case applic subscrib subscript inform subscript arriv termin api shown figur filter primarili callback procedur call match data arriv oper attribut filter direct access messag includ identifi previou destin node node ucb figur oper platform handl const int handl int int figur basic api handl int void handl void handl figur filter api evalu addit level control optim geograph inform avoid flood exploratori interest expect interfac extend gain experi filter inform requir api design favor program successfulli environ target program avoid synchron error avoid memori perform overhead multithread evid grow softwar suit platform filter applic run memori address space core space oper small sensor node lack memori protect perform optim multipl address space desir robust isol filter applic declar rout dan defin basic api develop independ implement lincoln declar rout system principl applic depend filter run implement level portabl demonstr queri proxi run implement declar rout data similar data differ rout transmiss applic core system primari differ declar rout includ filter applic directli influenc filter critic compon enabl gener data process lincoln declar rout includ direct support energi rout rout select avoid node gener move target geograph area current implement interest exploratori messag flood network gradient set direct commun explor filter optim geograph inform subset approach implement small processor distinguish extrem small memori footprint complementari approach full system subset full retain attribut singl support limit filter result add byte code byte data host oper system full system requir daemon static size librari implement compon add code data support radio photo entir system run memori static configur support activ gradient cach packet relev byte packet requir port requir chang applic accommod case implement strict attribut match reduc logic header format compat full implement implement softwar gatewai implement provid filter essenti compon enabl aggreg plan add intend leverag abil air program filter dynam region sensor distribut photo sensor room detect chang light temperatur sensor fine grain provid sensor data process abil commun node provid addit capabl wireless commun condit envisag tier architectur larger smaller node node form highest tier act gatewai tier tier connect sensor run network program tier node control filter program capabl node implement discuss draw observ experi implement rang implement suggest idea code portabl independ implement main declar primari implement run multipl platform june port progress radio requir modest term cpu speed processor memori megabyt support radio bandwidth radio design packet size small requir moder size packet code fragment demonstr implement subset processor common fulli custom protocol observ suggest preclud due size complex applic techniqu sensor network applic techniqu detail techniqu illustr process build effici applic sensor network approach examin data process reduc data traffic conserv energi approach provid nest queri sensor cue briefli describ applic implement data aggreg anticip sensor applic queri field sensor action sensor activ surveil system notifi biologist anim enter region coverag deploi sensor overlap ensur robust event trigger multipl sensor sensor report detect commun energi cost reduc data aggreg return user data aggreg binari area detect quadrant aggreg infrar sensor chanc detail aggreg common system problem design mechan establish data dissemin path sensor aggreg implement kind data fusion tradit network node order determin sensor present bind servic exist geograph list node identifi sensor region sensor elect algorithm dynam elect network node aggreg data return result architectur realiz opportunist data aggreg sensor select task achiev node geograph attribut data sensor intermedi sensor return path identifi cach relev data achiev run filter intermedi node suppress duplic data simpli propag slightli delai aggreg data multipl sourc experi influenc dynam select aggreg point minim data movement opportunist data aggreg benefit aspect approach filter provid natur approach inject code network attribut match filter remain inact trigger relev data common attribut set mean filter incur network cost interact directori map servic prior work analyz perform aggreg simul section evalu implement real sensor node valid initi result laboratori test nest queri event occur chang person enter room correl light open presenc absenc sunlight sensor network correl trigger secondari sensor base statu effect nest queri insid reduc duti cycl sensor reduc energi consumpt secondari sensor consum energi initi acceleromet trigger gp network traffic trigger imag gener traffic constant video process applic resourc motion sensor trigger user user figur approach implement nest queri squar initi grai circl trigger larg circl user thin dash line repres commun initi bold line commun trigger sensor figur show approach user sensor trigger network case assum sensor locat node commun directli part show direct implement user queri initi sensor sensor user queri trigger sensor small grai altern shown part approach user queri trigger sensor initi sensor nest queri approach grew discuss philipp bonnet queri optim advantag nest queri data initi sensor interpret directli trigger pass user monitor applic initi trigger sensor close cover physic user distant nest queri local data traffic trigger event send distant reduc network traffic latenc network typic reduct latenc reduct aggreg bandwidth user differ overload oper network challeng nest queri robustli match initi trigger sensor select good trigger sensor desir implement direct queri straightforward sensor user subscrib data initi sensor detect request statu trigger sensor subscrib recent direct queri illustr util predefin attribut identifi sensor type make geographi optim rout nest queri implement enabl code trigger sensor watch nest queri code relev initi sensor activ local trigger sensor demand multipl trigger sensor accept reason definit central select tion algorithm algorithm trigger sensor nomin random delai inform peer locat elect approach inspir srm repair timer peer disput claim locat extern frame refer defin node timer weight distanc minim number disput claim section evalu nest queri experi testb applic addit approach explor system research effort research cornel system provid commun applic repres visual sensor field queri proxi sensor node applic attribut identifi sensor run queri proxi pass queri proxi origin idea nest approach nest queri futur work includ understand network inform queri optim altern approach nest queri research bae system pennsylvania state system collabor signal process bae system contribut signal process code system provid sensor fusion algorithm combin system system commun data sensor data time filter architectur interest futur work evalu sensor fusion filter evalu approach paper effici implement improv distribut system sensor net section applic emploi techniqu measur benefit aggreg nest queri verifi raw match perform aggreg benefit section argu easi build sensor network applic filter earlier observ aggreg import perform data valid result actual implement simpl surveil applic filter examin aggreg testb sensor node distribut floor isi sensor connect rpc modem radio provid attenu antenna commun confin space exact topolog vari depend level rf network typic hop evalu effect aggreg sink side topolog node data sourc side node typic hop sourc gener event repres detect figur node posit sensor testb light node remain dark node floor radio rang vari greatli depend node longest stabl link node object rate event experi repeat event artifici physic sensor signal process event gener byte messag sequenc number synchron experi start node configur aggreg filter pass uniqu event suppress subsequ event ident sequenc number scenario abstract detail complet sensor network real signal process captur essenc network compon aggreg compar energi expend receiv event measur directli reason hardwar directli measur energi consumpt run system observ choic mac protocol complet domin energi measur low power mac protocol sleep period domin amount time spent choic protocol protocol sensor network experi mac approach measur energi consumpt estim effect reduc traffic mac duti cycl simpl model energi consumpt ô ø defin rel power time spent send defin requir listen duti cycl fraction time radio listen receiv traffic destin found sensor network pocket sever radio time rel energi oper sensor network timestamp sequenc number requir time synchron global gp sequenc number time experi synchron clock synchron sequenc number timestamp byte receiv distinct suppress suppress number sourc figur byte normal number distinct vari number sourc consumpt measur ratio assum energi consumpt ratio energi usag node duti cycl complet domin energi spent listen duti cycl half energi spent listen duti cycl begin domin send cost duti cycl radio todai radio node duti cycl analysi illustr import mac protocol directli measur energi figur measur byte node system normal number distinct event receiv point graph repres experi confid interv perform sourc basic ident suppress form suppress requir data event multipl sourc experi suppress suppress amount traffic roughli constant number sourc data aggreg show benefit process show traffic repres data control traffic compar traffic suppress show suppress reduc traffic sourc network exhibit high loss rate level traffic current mac perform simpl carrier detect lack messag broken loss singl fragment result loss hidden termin mac perform poorli high load work mac protocol confirm result simpl traffic model approxim messag long add interest messag flood reinforc messag reinforc path sink simpl data messag data reinforc aggreg exploratori data messag data sourc flood turn possibl data messag sourc incur cost full data messag aggreg hop incur hop cost aggreg point messag travel sink sum messag cost normal event expect aggreg provid flat independ number expect byte event aggreg number sourc rise shape predict match absolut term aggreg case differ due experi tend drive middl event gener experi deliv congest portion experi sourc high traffic normal fewer event sourc happen data sourc aggreg amount data lower predict observ longer path experi expect measur aggreg valid previou simul experi wider rang scenario previou simul studi shown aggreg reduc energi consumpt factor larg network activ sourc sink care compar energi byte energi save sourc greater traffic save observ sourc primari reason differ differ ratio exploratori data messag system exploratori messag rate messag select good gradient flood node data messag messag reinforc gradient form path sourc sink simul ratio exploratori data messag sourc messag data messag model testb ratio messag data messag roughli ratio experi small radio bandwidth reason run time larg differ ratio consist larg differ energi traffic save potenti disadvantag data aggreg latenc effect aggreg latenc strongli depend aggreg algorithm algorithm experi affect latenc forward uniqu event immedi recept suppress addit duplic addit neglig cost search aggreg delai transmit sensor read hope aggreg read add latenc understand aggreg sensor fusion algorithm import area futur work quantifi benefit aggreg specif aggreg process exampl rang simpl data audio event deliv number light sensor figur percentag audio event successfulli deliv user cach collabor signal process experi attribut match filter make aggreg similar servic easi enabl notic perform improv nest queri benefit section suggest nest queri reduc network cost argu nest queri implement attribut filter valid claim potenti perform benefit implement measur perform applic nest queri applic similar section figur user request acoust data correl light sensor testb shown figur place user node audio sensor node light sensor node hop light sensor audio hop user node provid reproduc experi simul light data chang automat minut minut light sensor report state special attempt made synchron audio sensor gener simul audio data time light sensor state light audio data messag byte long figur show percentag light chang event successfulli result audio data deliv user point repres experi show confid total number event number time light sourc chang state success event audio data deliv user deliveri rate reflect messag deliveri rate cumul effect send data hop nest flat system primit mac protocol impact congest miss event translat detect latenc sensor network afford miss event retransmit time sensor loss rate unaccept high set interest set data class interest class data task eq task confid gt confid latitud ge latitud latitud le longitud longitud ge target longitud le target figur attribut match experi tional system experi sharpli contrast bandwidth requir nest flat queri sensor flat queri show significantli greater loss nest queri light audio data travel user flat nest queri suffer greater loss sensor queri fall compar deliveri rate nest queri queri show local data sensor import parsimoni bandwidth network expect nest queri oper lower level data traffic queri lower radio duti cycl longer network lifetim cost match attribut match commun applic system technolog trend suggest rapid improv processor sensor node chose hold perform constant leverag technolog reduc price perform consid constraint memori small implement evalu match perform examin cost match data sensor basic match case compar interest data shown figur evalu cost larger data object number attribut data attribut experi sensor node cpu evalu cost singl match measur cost match match loop repeat experi time avoid undu system effect interrupt order attribut set random experi show confid memori cach approach underestim cost basic trend identifi applic oper system expect cost match linear number element confirm figur show cost match number attribut attribut set wai lowest line show case attribut set match set confid set chang match algorithm test formal set increment cost addit attribut set fairli small insensit type attribut ad fail formal set expect cost higher match number attribut set figur match perform number attribut grow measur higher line show cost match attribut succe differ cost addit attribut line show cost addit match line addit attribut formal eq attribut match set addit attribut actual examin requir search current implement complet absolut perform oper reason small attribut set slow match set slow internet router reason sensor network expect event happen frequenc measur suggest potenti optim match perform segreg actual formal reduc search time formal match formal compar attribut static dynam optim move attribut match front plan explor kind optim futur experi discuss experi provid insight sensor network build substanti prior simul studi experi examin nest queri match perform suggest cpu overhead match constraint power sensor node nest queri greatli reduc content local data movement experi explor oper previou simul studi sensor network radio actual hardwar protocol scenario behav qualit sensor network common wireless lan earli oper experi demonstr difficulti oper sensor network high util observ suggest area futur sensor network adapt local node densiti begin explor area work understand paramet map overhead reliabl present frequenc exploratori reinforc applic oper open feedback congest control aspect radio propag prove unexpectedli difficult experi show asymmetr link fine direct poor imposs work asymmetr link provid intermitt connect futur direct send similar data multipl path gain robust face link current simul statist adequ reflect observ propag characterist gener happi approach attribut filter easi build adapt sampl applic debug softwar futur work work describ current approach construct robust distribut sensor network applic suggest area futur work includ enhanc testb appli addit understand build sensor network plan testb hardwar plan move radio rf monolith ucb packet control control rpc rapid approach give complet control mac protocol explor perform simul testb experi aggreg show qualit result evalu step experi parametr simul work repeatedli challeng difficulti understand network dozen physic distribut node current environ augment radio network separ wire network data work develop analysi tool network tool report chang radio observ rate energi permit flexibl accur synchron node clock begun work monitor tool work mac protocol sensor network continu challeng spite publish work area ongo freeli energi awar mac protocol remain explor altern hope solut forthcom balanc control data traffic import system sensor network techniqu constrain control traffic exist protocol wire network approach appli system explor applic sensor network collabor applic remain interest direct explor collabor signal process interact process filter focu wireless sensor techniqu develop relev wire sensor network wire connect greatli reduc bandwidth constraint elimin power reduc system complex decoupl data sourc process reduc latenc improv scalabl prior system separ abstract virtual inform futur direct appli wire sensor network coupl physic world paper approach distribut system built data process attribut extern mean sensor type geograph lowest level approach avoid multipl level bind common approach data turn enabl process support data nest queri similar techniqu critic reduc network traffic conserv energi evalu effect techniqu quantifi benefit process data aggreg nest queri experi found aggreg reduc traffic nest queri reduc loss rate aggreg studi experi evalu techniqu oper testb approach import emerg domain wireless sensor network network power resourc constraint fundament acknowledg dan work defin network rout api van jacobson suggest appli direct sensor network peopl particip develop includ alberto jeremi lewi jerri zhao jerri zhao set debug station jeremi prompt modif rpc devic driver measur paper imposs input group uc berkelei support discuss paper benefit contribut numer includ philipp brian nobl paper sosp review refer design implement intent system symposium oper system page dec acm katz activ servic framework applic multimedia transcod acm sigcomm page sept acm mobil network ieee person commun oct birman process group approach reliabl distribut comput commun dec seshadri queri process devic system technic report cornel oct peterson reason system acm transact program languag nov perform wireless ad hoc network rout protocol intern confer mobil comput page oct acm brook iyengar robust distribut comput algorithm ieee june carriero gelernt linda kernel tenth symposium oper system page dec overview model servic recommend zhao habitat applic driver wireless commun technolog acm sigcomm workshop data commun latin america san costa apr acm hong distribut anonym inform storag retriev system icsi workshop design issu anonym juli silva network rout applic interfac walk dec declar sensor network spie integr command environ san juli spie spie intern symposium optic scienc katz architectur secur servic discoveri servic intern confer mobil comput page aug acm estrin transact identifi dynam sensor network intern confer distribut comput page apr ieee floyd jacobson resourc manag model packet network transact aug rout address assign aggreg strategi rfc internet request sept balakrishnan commun protocol wireless network hawaii intern confer system jan balakrishnan adapt protocol inform dissemin wireless sensor network intern confer mobil comput page aug acm system architectur direct network sensor intern confer architectur support program languag oper page nov acm imielinski goel queri monitor deepli network collect physic space ieee person commun special issu smart space octob estrin direct scalabl robust commun paradigm sensor network intern confer mobil comput page aug acm jacobson compress header serial link rfc internet request feb jacobson adapt web global cach architectur intern world wide web june domain facil rfc internet request nov inform architectur extens distribut system symposium oper system page north dec acm peterson servic network acm sigcomm confer page aug kaiser wireless integr network sensor commun unicast address format rfc internet request jan jacobson scalabl timer soft state protocol ieee apr ieee singh power awar protocol signal ad hoc network acm comput commun juli sensor network confer sept survei activ network research ieee commun jan waldo jini architectur comput commun oct zhang refresh ieee intern confer network oct ieee estrin energi conserv ad hoc rout intern confer mobil comput page juli acm kill lightweight directori access protocol rfc internet request mar govindan geograph rout wireless sensor data dissemin protocol work mar estrin residu energi scan monitor wireless sensor network technic report 
Agents/2227 1 find move constraint cyberspac hui preec depart comput scienc king colleg aberdeen aberdeen scotland unit kingdom abstract architectur effect method construct distribut inform system kraft system exploit exchang inform form constraint data particip agent kraft approach solv design configur constraint data retriev agent repres custom vendor extranet transform common process mediat agent paper describ kraft discuss issu involv join kraft network inform provid examin role autonom mobil agent kraft introduct tradit distribut system provid uniform transpar access data object network data semant knowledg store type store knowledg constraint state constraint abstract constitut import part semant data model interest declar quantifi constraint abstract repres partial solv solut intermedi result mobil knowledg export attach constraint restrict data form relationship object constraint combin manipul distribut environ constraint highli suitabl knowledg distribut applic wallac enabl approach solut design configur problem constrain tutor constrain residu figur exampl demonstr grai express constraint student record constraint languag applic domain protein structur restrict bond length background relat work kraft system al constraint knowledg held distribut transform ontolog quantifi constraint grai figur gener declar form predic definit comput function effect recip select calcul pass fuse transform recip heart kraft architectur open possibl architectur prove effect approach develop distribut inform system al support rich knowledg reason content open environ resourc join leav network dynam genesereth kraft emploi architectur al provid requir extens adapt dynam distribut environ unlik distribut inform kraft exchang data constraint agent system design kraft architectur build recent work distribut inform ua ua ua ua user agent resourc wrapper facilit mediat kraft facil compon kraft domain kei auxiliari figur conceptu view kraft architectur tem role identifi kraft agent similar infosleuth system al infosleuth primarili concern retriev data focu kraft combin data constraint kraft build work knowledg share effort al facilit brokerag method subset kqml specif unlik kse attempt support agent commun rang knowledg languag attend translat kraft view constraint good express tractabl emphasi kraft similar xerox constraint base knowledg broker project differ kraft transform constraint extract local typic reason ontolog schema mismatch al al system architectur user agent resourc kraft interfac kraft domain wrapper wrapper provid translat servic intern data local user agent resourc data integr schema act share kraft domain resourc incap handl asynchron wrapper handl buffer schedul request resourc kraft user agent serv infosleuth al user agent kraft wrapper call resourc agent infosleuth terminolog facilit provid intern rout servic messag kraft domain maintain directori kraft locat servic detail load reliabl princip function accept messag kraft facil rout appropri facilit provid rout rout messag partial address wholli base content servic requir kraft facilit correspond infosleuth broker agent kqml facilit primari focu kraft level aim provid rang mediat manipul knowledg form constraint problem decid catalogu order compon constraint extract mediat gener descript compon conjunct constraint translat domain ontolog design ontolog mediat result collect constraint transform singl constraint repres fuse exist set design constraint constraint solver mediat coordin search solut make solv resourc kraft mediat compar task execut agent instanc wiederhold mediat concept genesereth configur problem obviou area applic configur problem tackl system famou configur comput nowadai tackl constraint satisfact problem kraft architectur domain variabl entiti store remot constraint entiti type set store constraint found mediat pass constraint solver restrict solver find feasibl valu satisfi common engin problem note problem complic constraint refer relat instanc entiti valu extract check compat configur problem solv special written piec softwar includ preprogram constraint paramet valu number flat file prepar design kraft architectur paramet constraint repres problem search select brought network node develop unanticip wai agent architectur hope cope evolutionari chang autonomi resourc node kraft problem find number part fit make work supplier part make form internet tabl semant hidden assumpt assumpt asterisk footnot small print part mount adequ size make distribut queri find list ensur part satisfi constraint knowledg constraint aim transform work context share ontolog integr data constraint store metadata constrain widget constraint express kraft constraint interchang format version base languag express semant object grai constraint repres form trigger frame translat constraint kraft network make widget translat constraint form consist share ontolog requir understand terminolog share constrain wai transform transform fuse constraint usag part constrain altern repres fuse constraint collect normal form fuse inform wai explain role constraint kraft constraint excel declar semant featur data model import abstract extend data model wai address question import era internet chosen languag develop function system base shipman languag origin integr data express local local schema found constraint languag independ problem domain repres knowledg store varieti local data model power order logic safe express restrict mix quantifi finit domain object store finit subrang integ power function program languag comput constraint kraft sourc kind constraint store associ class descriptor data view attach instruct data object repres small print condit refer earlier data object annot declar knowledg transform process constraint gener applic problem data data object attach instruct extract satisfi ensur data properli note constraint store class descriptor select condit constraint tutor make specif object type constraint repres domain specif problem solv knowledg constraint solut store result satisfi applic problem constraint specif applic problem potenti candid satisfi order qualifi solut initi solut hold actual data framework problem solv knowledg solv popul solut satisfi constraint constraint repres restrict user specif requir solut resourc serv inform sourc feed knowledg system form constraint user requir specif applic problem instanc user queri solut constraint fuse mediat user agent data object constraint problem specif problem solv knowledg constraint satisfact problem solut solut constraint solver figur diagram show constraint fragment export resourc fuse mediat constraint satisfact problem kraft search fusion process solv design configur retriev data object popul solut satisfi integr constraint attach solut relev data object role constraint evolv state portabl effici approach achiev export constraint fragment constraint fuse mediat descript constraint satisfact problem configur task al csp queri constraint logic program fed distribut constraint control mediat wai fuse check constraint set object retriev distribut queri reject satisfi condit combin select inform constraint refin distribut effici principl semant queri constraint logic solv techniqu grai hentenryck complex set interlock form solut applic problem kraft csp csp distribut constraint solv system kraft csp involv simpl constraint huge number candid data object multipl requir strategi focu data filter comput effici constraint fuse demonstr constraint fusion configur problem pc built combin compon vendor user specifi requir form constraint specifi pc processor constrain pc compon fit satisfi constraint solut size os smaller equal hard disk space proper constrain pc candid compon vendor instruct attach constraint vendor oper requir memori constrain pc fuse constraint descript constrain pc true issu mobil agent constraint kraft network architectur appli problem gather specif configur includ potenti part constraint target extranet vendor potenti custom form virtual privat network share inform order prepar conform map data ontolog share monoton extens al cost join kraft network section requir kraft approach particip agent main issu conform data model role agent autonomi constraint data model wai data model make lack full natur languag comput agent exchang messag natur languag partit data valu object distinct data associ distinct semant natur languag concept describ partit vocabulari attribut relationship appli object class comput mean check term concept interchang fulli understand mean natur languag valid appeal correspond instanc object object list oper combin data type typic import part data model defin oper relat algebra relat term predic logic lambda function approach fact defin data structur model includ kind model case tool design softwar construct describ describ invari preserv state classic role integr constraint level semant point set rule calcul combin set data heterogen produc result give data label give mean notion understand human replac set rule oper comput meaningless fulli grasp familiar approach type checker program languag point mylopoulo repres semant extern insid comput engin distinguish er model underli ontolog semant terminolog superfici resembl work kraft grai convinc data model order express constraint establish finit domain real world entiti basi quantif attribut method appli type data method entiti type state constraint includ aspect entiti data typic work restrict view aspect interest internet web page natur languag hard extract constraint automat check make mistak writer sentenc consist make possibl mismatch greater hope present xml standard enforc vocabulari integr schema section document web textual inform small number fact tabl section conform larg data similar success approach extract inform html tabl web site constraint autonom agent kraft system agent resid processor job wrapper locat found facilit queri involv collect data number locat perform mediat send number asynchron request step inform put combin follow techniqu process distribut select inform implicit kei tupl return queri provid extra select condit queri inform join conjunct approach kraft mediat plan gener queri cooper constraint base metadata constraint type system agent imagin set number agent parallel strategi visit site search data found send number peopl treasur collect object sequenc situat agent accumul data constraint resourc search progress constraint act bit clue treasur agent modifi plan conjunct queri genesereth strategi interest solut quickli suitabl systemat compar solut find collect data batch process techniqu economi scale improv perform depend constraint conjunct disjunct pai select constraint agent shop hook larg spend lot time collect relat data find arent restrict select case approach plan win search data satisfi disjunct advantag agent radio search result back headquart broadcast agent abandon search concentr work report concentr transform satisfact data intens data model describ store relationship object distribut knowledg base relev agent domain variabl constraint restrict subrang integ object belong class object entiti type er model survei articl agent constraint wallac begin note natur synergi strongli ndumu classifi interest inform internet constraint agent concentr gener agent acquir vari proce visit usual agent hold constraint advantag work function model effect treat function prolog implement languag concentr kraft architectur collect constraint solv configur tend spatial tempor schedul problem agent appli sycara kraft prohibit collect constraint schedul experi develop agent kind solver unusu featur architectur includ facil transform constraint express local ontolog conform share ontolog aspect integr order constraint express common basi combin crucial set configur problem work constraint depend local data model map share ontolog easili join network map export knowledg share ontolog easi hope agreement label xml denot part web includ tabl conform vocabulari ontolog specul consequ extend architectur mobil agent address al falt interest issu concern tradeoff agent behav behav plan coordin fashion send constraint back mediat constraint solver plan coordin search data satisfi constraint ai studi move constraint singl address space solv complex plan problem extend collect move constraint order set csp applic agent acknowledg acknowledg support bt kit hui work kraft support epsrc graham kemp cui kraft project partner interest discuss figur paper workshop grai refer constraint agent decentr network infrastructur constraint paper aaai menlo aaai press constraint agent inform ag journal comput scienc function constraint languag implement data knowledg engin semant integr inform open dynam environ sigmod knowledg base semant comput theori inform direct system ablex publish corp inform system reason structur collect text intern confer autonom agent mn acm press constraint confront ignor ai magazin compil languag semant integr constraint ifip work confer data chapman hall plan complex updat satisfi constraint rule constraint logic search engin proc int workshop rule system lectur note comput scienc declar express semant integr protein structur data manag intern workshop inform technolog san ieee comput societi press knowledg fusion distribut knowledg base wagner system handl distribut design constraint workshop manufactur dept comp scienc univ minnesota constraint data fusion distribut inform system advanc british nation confer springer verlag semant agent commun languag phd baltimor usa distribut problem solv coordin societi agent knowledg ifip work confer design construct enabl technolog knowledg share ai magazin introduct agent technolog function data model data languag acm transact system order conjunct queri artifici intellig plan constraint artifici intellig java constraint bring constraint technolog internet java languag constraint paper aaai menlo aaai press van constraint satisfact logic program mit press analysi ontolog heterogen versu aaai spring symposium ontolog engin ed eighth intern workshop expert system applic ieee comput societi press basi mediat intern confer cooper inform system 
Agents/62 1 institut ur informationssystem system technisch universit wien austria institut ur informationssystem system agent thoma eiter mascardi subrahmanian research report octob research report research report octob agent thoma eiter mascardi subrahmanian abstract agent todai internet world expand rapidli agent develop proce larg optimist assumpt agent error agent numer reason agent share workspac agent human updat made entiti agent face situat explicitli program deal error code agent lead inconsist situat unclear agent act defin agent execut model agent continu act error type occur agent action automat confront repair continu engag work interact agent unaffect repair institut und ludwig wittgenstein labor fur technisch universitat austria universita di itali part work visit tu wien institut advanc comput institut system research depart comput colleg maryland work support part austrian scienc fund project armi research lab contract number nsf young investig award grant number aro grant grant copyright author rr content introduct impact preliminari architectur formal definit architectur corrupt code call atom suspici code call atom agent state repair repair data structur repair action librari error toler agent cycl method implement suspici add implement implement implement relev agent framework kowalski sadri unifi agent architectur rao georgeff bdi architectur wooldridg comput system relat work inconsist agent framework reason statu set safeti rr introduct agent rapidli grow area research artifici intellig rang span server web search engin numer paradigm agent ai literatur past author work framework call impact maryland platform agent collabor develop theori exist legaci code base data sourc agent state resid data structur messag agent state agent action accord oper principl ensur result state satisfi integr constraint exampl state includ receipt clock receipt servic receipt servic updat data eiter show strong connect agent theori classic method logic nonmonoton reason show shoham aop orient system larg simul impact larg part architectur captur framework framework fact agent decid state major problem address agent framework includ impact assum rule suffici appropri respond request arriv assumpt agent develop cover optimist unreason assumpt program program question agent confront situat act case legaci note legaci system exist gui agent access updat data legaci gui alter agent state wai agent find unaccept agent corrupt extern entiti agent current state violat integr agent unabl find set action execut current state code tackl problem consid extent nonexist statu set integr constraint violat paper present architectur algorithm agent exhibit import properti recoveri agent recov continu agent continu process request continu recov import agent servic lot request organ paper section present overview impact agent architectur add compon error recoveri compon architectur section section provid formal set wrt semant agent assum ration reason statu set semant agent rr legaci data function call action polici secur action base constr integr action constr messag messag agent figur basic architectur impact agent definit request affect agent corrupt unaffect request continu process corrupt corrupt agent attempt recov section describ special repair data structur repair action recoveri compon select repair action host realiz repair section discuss agent result tool previou recov error show impact agent recoveri present modif agent cycl incorpor desir properti recoveri continu section discuss work appli agent framework kowalski sadri bdi work wooldridg relat work discuss section direct futur work discuss section impact preliminari applic program reason type program deal type data manipul varieti critic notion agenthood applic arbitrari softwar program agent develop select data structur suit applic function desir user applic build figur show architectur impact softwar agent import note agent architectur content compon lead behavior capabl offer agent agent data structur agent built exist bodi abstract definit bodi code specif data type data agent manipul data type domain space object type data type countri type countri instanti content data type subset space type rr set data structur manipul set extern program function constitut applic programm interfac api packag top agent built agent includ specif signatur api function call type input function call type output function unifi languag queri softwar packag leverag function defin term code call code call function defin packag state list code call argument valu assum set singl object coerc set code call atom express form term cc code call ground succe answer set valu return fail answer variabl return result answer set ground substitut return true uniform view ground case identifi answer true set void substitut answer empti set substitut code call atom denot logic negat code call extend natur set code call atom xg code call condit conjunct code call atom constraint involv oper constraint atom access field variabl rang record field check state condit constraint atom form op op term code call condit provid power syntax access heterogen data structur code call condit oracl imag complex condit join data oracl imag select peopl make oracl find pictur person person call mari gener notion join relat join relat imag agent assum access messag box data api function call access detail messag box impact found point actual set object data structur messag manag agent constitut state agent identifi state set ground code call true action agent set action variabl chang state action includ read messag messag respond execut clone copi agent move remot updat agent data action express list term action atom repres set action result variabl instanti valu action execut agent action rr precondit code call set effect add list delet list code call describ agent state action execut method implement program languag script languag user deem consist bodi physic code implement action notion concurr agent bodi code implement notion concurr set action current agent state return singl action input action output notion concurr properti state restrict code call atom occur add delet list action make assumpt paper action constraint agent finit set action constraint rule form state satisfi code call action concurr present disregard action sine easili elimin integr constraint agent finit set ic integr constraint ic state agent satisfi ic resp form code call code call atom constraint atom ic mean statement function depend relat packag db express integr constraint db db return tupl relat assum integr constraint exist agent state satisfi integr constraint ic set integr constraint consist determin inconsist undecid autom check softwar agent legal initi state state collect state satisfi integr constraint specif integr accept hold agent program agent set rule call agent program principl agent oper rule deontic agent express action call action statu atom action statu atom read oblig action statu call action statu liter agent program finit set rule action statu code call action statu liter due space repeat semant agent program overview appendix detail architectur formal definit discuss extend architectur fig handl case agent error due violat integr valid statu agent state autonom updat parti omit place safeti aspect appendix rr architectur assum mechan notifi agent state chang parti assum abstract agent receiv messag agent agent answer code call atom resp term cc ground agent agent answer code call atom previou form code call atom set ground substitut mean parti updat agent state answer chang answer minu substitut substitut error occur agent situat incom messag form trigger error valid statu set incom messag entiti send agent messag form updat violat integr constraint leav state invalid deal situat agent develop build impact perform task order impact agent recov corrupt set ra repair action properti section object function evalu cost state idea agent code repair compon automat repair action comput state satisfi integr constraint gener valid statu user specifi compon agent section specifi impact agent develop environ automat convert agent compon repair compon execut bodi java bytecod deploi continu properti agent preserv requir agent state corrupt action extern agent continu process request servic long request ongo repair corrupt part agent manag relat relat find extern corrupt relat queri access relat process agent corrupt relat repair reader messag error incom messag agent chang agent state messag updat agent messag box integr constraint violat messag agent develop write patent absurd integr constraint syntax ic agent develop write rule messag box messag agent messag box messag agent agent corrupt messag agent arriv problem avoid ad restrict syntax agent program type regular agent program introduc avoid problem addit requir ic mention code call atom involv messag allevi problem rr proce section address problem agent code call atom resp set code call code call atom potenti corrupt method appli base syntact analysi agent integr constraint introduc section notion code call atom determin agent make affect potenti corrupt code call central recoveri section corrupt code call atom set code call atom code call atom integr constraint affect defin procedur call set code call atom integr constraint corrupt preliminari definit introduc notion subsumpt code call atom definit call set code call atom subsum set code call atom written instanc substitut singleton set omit write rest implicitli assum code call atom standard unif code call atom subsum defin associ code call condit set code call atom set code call atom occur definit atom set code call condit code call atom set induct defin code call atom constraint code call condit integr constraint ic defin defin describ set potenti corrupt code call atom code call atom corrupt definit integr constraint ic code call atom unifi gener unifi current result code call atom lead inconsist integr constraint code call atom turn corrupt extern entiti modifi state due section rr consid code call atom occur consid corrupt notic unifi gener unifi easili nest term integr constraint ic defin procedur set code call atom consid set code call atom consid corrupt proc set code call set code call ic return end proc notic implement inflationari oper set code call termin finit input output compact remov subsum code call atom reason iter appli corr oper procedur error mask integr ic ic ic ic current state code call atom current state ic violat potenti evalu return result violat integr reflect condit current state coher agent set integr constraint violat current state potenti correct evalu true fact correct evalu case corrupt evalu order satisfi ic rr violat exclud possibl problem propag safe integr constraint corrupt potenti corrupt code call integr constraint ic ic violat current reason suspect code call atom corrupt complet unrel corrupt atom sound approach express proposit state coher state reach chang return valu corrupt code call maintain return valu agent state set ground instanc integr constraint ic violat state set code call atom proposit set code call atom exist agent state ic ground code call atom differ mean turn modifi return result corrupt code call chang result code call atom proof defin suitabl recal agent state exist satisfi integr arbitrari agent state ground code call atom defin notic differ ground subsum ic ground instanc integr constraint ic case exist definit hold iff ic hold impli ic conclud iff ic case ic prove result continu previou state code call atom true code call atom consist reach simpli chang return implement cautiou larger set code call atom corrupt semant appli case case refin pictur minim set code call identifi corrupt view reach coher state chang valu code call atom comput minim set code call atom chang lead gist result rr theorem set gi ground violat ground integr constraint current agent state ground code call atom decid smallest set ground chang valu consist state result proof variant satisfi problem reduc problem fc cm set proposit atom negat softwar packag maintain truth assign proposit api return variabl set true distinguish atom assign true satisfi iff atom agent state atom set integr constraint integr constraint violat belong smallest chang ground code call atom turn consist state iff satisfi assign decid gi easili construct polynomi result suspici code call atom chang subset ground instanc corrupt code call atom recov agent state agent cycl repair procedur repair arriv simpli queu agent find process messag repair proce handl case introduc notion action atom notion code call evalu action atom affect repair access code call atom possibl chang repair process deontic statu action atom chang repair complet impact action atom deontic statu determin run agent program rule program involv affect action atom corrupt code call atom propag action atom code call atom bodi rule consid affect rule fire treat corrupt code call code call safe evalu current agent state affect corrupt state affect attempt repair corrupt part state atom safe evalu repair process agent process messag natur evalu code call atom process messag repair state long process messag delai avoid potenti incorrect result similar appli process messag form case corrupt code call determin suspici code call atom syntact analysi agent program defin procedur set code rr call atom subsum corrupt ground code call atom agent set suspici code call atom procedur oper phase determin code call atom corrupt backward propag integr constraint violat complet repair defin direct action atom code call atom definit affect action action atom directli code call atom exist unifi mgu directli affect directli directli statu evalu ground instanc involv overlap ground instanc code call atom precondit chang repair defin action atom set affect action code call atom definit rule action set action atom occur set action code call unifi mgu directli unifi mgu set union set union rule affect defin exist set action atom action atom assum code call atom corrupt subsum coarser view ground atom simplic effici remark semant appli agent program definit refin case reason statu set semant action atom ad bodi agent manag polici depart store classifi custom medium low classif send custom practic sophist classif rule agent program oracl rule custom enter assum custom high salari del list custom rr add exampl pair oracl code call atom unifi oracl mgu fc oracl directli code call atom unifi oracl mgu fc action atom unifi head mgu fc notion determin action atom affect build statu set agent depend semant wai includ action statu atom statu ration reason statu set action statu atom belong statu set occur deriv action deontic closur rule def feasibl statu set includ case occur respect assum program dummi rule action rule easili ad chang semant program posit defin comput set suspici code call atom set code call atom suspici procedur defin tabl suspici code call atom determin phase iter determin code call atom affect syntact examin rule agent program start knowledg code call atom input algorithm corrupt code call atom bodi rule found affect suspici end phase code call atom possibl affect corrupt code call atom determin lead agent action vari dramat agent develop origin code call atom unintend consequ address corrupt code call condit trigger unintend action care account fact action action integr effect occur integr constraint joint execut corrupt repair state longer execut execut illustr add list code call atom add list integr constraint ic rr proc set code call set code call subsum corrupt ground code call atom phase propag begin find action atom troubl ic ic ic unifi mgu unifi mgu phase back propag poss atom begin return affect action atom return bg end proc tabl procedur rr assum current execut ic integr constraint involv corrupt code call atom repair agent execut execut simultan ic violat repair agent comput statu set execut mean statu action agent corrupt state repair undesir elimin procedur comput action atom lead problem phase comput action atom deriv action atom analyz rule program atom occur mean unifi action atom bodi rule suspici code call atom rule bodi consid effect result code call atom relev consid earlier phase phase code call atom affect rule contribut deriv problemat action atom back care fact action atom code call evalu comput statu set code call consid suspici found affect simpl agent ag integr constraint ic agent program consist ration statu semant add add told find suspici code call atom inform call iter augment initi set reach fixpoint rule fx add action atom set remain code call atom bodi occur rule affect occur action atom ad remain unchang rule affect rule newli affect occur fx unifi atom head newli ad ad rule newli rule affect rr iter bring phase termin comput delet list occur ic set line action add subsum action obtain phase rule affect note rule affect action statu atom unifi action statu atom affect ad affect ad affect unifi head fx sg ad ad affect loop return evalu ad precondit action statu atom omit subsum code call result theorem state procedur input set code call atom set code call atom arbitrari code call atom unifi code call atom base affect ongo attempt repair code call atom action result state involv isol corrupt code call state repair run agent program agent repli messag db queri tabl corrupt code call atom restrict db car tabl provid answer messag doesnt refer car preliminari definit agent program ground code call atom influenc set smallest set action action directli affect action ground instanc rule influenc set arbitrari code call denot union ground instanc theorem agent state repair set code call atom code call atom unifi valid statu set disregard exist valid statu set wrt ic iff hold modal op proof statu set exist lead state hold action belong ground code call corrupt code call subsum virtu phase instanc subsum rr contradict hypothesi statu atom statu set comput access ground code call atom action statu atom op ground code call atom af set action instanti action atom set comput defin statu set af affect action statu action show set lead state coincid result show feasibl statu satisfi condit def kei fact ground instanc rule satisfi af af satisfi rule clear satisfi rule condit satisfi ground action action statu atom belong feasibl statu clear satisfi condit case analysi yield ground instanc ic integr constraint ic satisfi assum corrupt code call code call ic action af chang code call action belong ic assum code call ic action af specifi chang code call ic phase procedur af action specifi chang code call atom ic ic ic code call atom ic chang action af ic violat state ic ic ic satisfi state condit satisfi show feasibl statu set wrt sem ration statu set show proper subset satisfi deriv contradict assum smaller action statu atom set action af smaller statu set satisfi state note ground instanc rule satisfi af deontic action consist action close ration statu set contradict hypothesi coincid wrt statu action af smaller wrt af statu set smaller statu set similar state mean ration statu set contradict prove ration statu set sem reason statu set show ration statu set reduct red reason statu set wrt feasibl statu rr set observ reduct preserv kei properti af similar argument obtain ground reason statu set wrt formal result assur case agent program admit singl statu set state true impact target class regular agent program action repli messag corrupt state repair agent state repair recal agent state content data structur order agent automat handl integr constraint violat lack statu add special set data structur agent call repair data structur data structur special api function call repair data structur repair data structur buffer consist messag wait servic involv access part agent state buffer consist corrupt code call atom buffer consist integr constraint undergo repair auxiliari buffer suspici code call atom set con state consist ground code call atom true distinguish consist state set curr state consist ground code call atom true current state repair data structur support api function code call atom return true impli set code call atom notion infer fix concret implement function wai implement check physic present check instanc code call atom check impli set axiom set implic rule procedur defin section rr add function code call atom corrupt repair return true return implement depend possibl exist section import note repair data structur api call includ part impact agent develop environ program agent agent develop repair action librari addit repair data augment agent set action agent set action agent state repair action implement straightforward extens dynam link librari provid impact agent develop environ definit agent state agent ra repair action librari agent ra iff ra iff state ra action ra execut execut yield state iff ra state ra reachabl state mean sequenc repair action transform illustr grid simpl scenario grid agent manag robot move repair action ra grid action move robot direct describ north similar assum underli softwar api function return posit robot time function call made north grid add grid grid state grid grid grid grid grid grid rr ra grid grid grid ra definit set ra repair action complet wrt agent state iff exist state ic ra complet wrt agent iff ra complet wrt state ra complet agent iff state agent execut repair action consist integr agent state obtain agent develop specifi repair critic complet wrt rest agent scenario previou scenario integr constraint exist state posit occupi robot ra grid complet wrt exist state ra grid satisfi integr constraint ra grid complet wrt agent agent robot occupi posit set repair action grid order provid system develop defin set repair action adopt context repair data structur introduc section state repair action ra init assum agent initi state init time init assum satisfi integr constraint set con init ra complet wrt init defin action ra defin ra repair curr repair con repair curr repair curr function curr con provid repair packag set ground code call atom true current set ground code call atom true distinguish state satisfi integr constraint section action ra appli assum retriev ground code call atom current state repair curr consist state repair con current state chang rr state repair action ra pref prefer state repair action agent develop initi con state pref satisfi integr constraint repair action ra roll instant agent track consist state lk set con equal lk function identifi proper repair action perform reach consist updat con state integr constraint repair agent state reset consist state set repair action consist singl action exampl content con dynam agent life cycl strategi usabl action agent execut fax action find imposs recal repair ra ic repair strategi appli condit integr constraint comparison atom head code call atom bodi agent integr constraint ic ng construct repair action ra ra add code call code call protect atom ra slight variant call ra includ list code call atom repair action ra similar ra delet list code call atom comparison atom prior system check integr constraint comparison atom head code call atom bodi result give idea difficulti check complet concret statement assumpt complex evalu code call domain data type assumpt make similar analysi complex agent program request size agent state bound polynomi size problem input ensur assum number argument code call bound number valu polynomi input code call agent state evalu polynomi time assumpt state refer theorem check complet set repair action ra wrt agent state assumpt undecid gener check complet set repair action ra wrt agent assumpt undecid gener rr proof pspace upper bound consequ fact size agent state bound polynomi pspace lower bound explain fact ture machin polynomi work space easili encod problem check complet set repair action respect agent harder check wrt agent state complet test wrt agent undecid shown reduc problem decid return true instanc relat undecid extens librari complet set repair action easili incorpor impact agent develop agent integr agent automat select ra init ra roll ra ic repair action strategi relev repair action automat comput fill agent impact agent develop environ error toler agent cycl solut problem agent recov corrupt state continu process request unaffect ongo repair note point agent state repair repair messag form attempt process request usual account possibl messag yield valid statu add messag start repair state agent state check ongo repair process current request check code call atom messag affect ongo repair add messag buffer process secur knowledg repair made agent state affect depend current code call atom note case state state repair affect incom captur condit state empti involv ongo repair case dealt uniformli simpl definit definit agent program agent state set wrt statu set wrt modulo condit feasibl statu set definit appendix show agent cycl modifi handl requir recoveri continu modifi agent cycl stand error defin tabl special procedur agent state assembl list ground action statu atom serial execut chang agent state satisfi integr constraint execut list buffer messag handl instanc violat integr repres set corrupt code messag rr proc agent msg msg msg ident dummi id true add affect msg state repair doesnt affect msg set wrt ic exist msg execut action statu set exist error condit set exist msg begin switch repair reiniti buffer instanc ic ic ic st ic ic add requir repair add add poss corrupt ic end msg begin ic ic msg set msg wrt ic exist begin add ic end comput set msg wrt ic execut action end end proc tabl modifi agent cycl rr express ic denot set integr constraint ground instanc integr constraint ic note level msg describ agent state updat state messag msg algorithm captur requir recoveri continu recoveri support step step handl case messag yield valid statu set extern updat integr constraint violat continu support step execut action safe theorem repair process messag step realiz current state repair process step realiz agent program se violat integr state repair state messag flush buffer process statu set comput method implement suspici add coupl altern wai implement function suspici add implement mention wai implement add suspici simpl import note add suspici mutual compat proc return true return end proc proc add end proc insert messag code call atom comput corrupt code call atom function defin section add start evalu suspici code call atom put auxiliari procedur advantag evalu simpl subsumpt check execut time proport product length tabl longest code call atom store disadvantag messag corrupt suspici code call comput insert space consum oper add expect agent state corrupt infrequ comparison number messag process concern corrupt agent state implement implement suspici add work code call atom insert code call atom corrupt comput suspici code call atom messag explicitli determin affect function rr proc return true return end proc proc add end proc unlik spend minor effort insert code call atom arriv attempt check request affect ongo repair find larg lot thing insert time check request affect ongo repair method worth number repair larg request implement procedur procedur function program languag terminolog side current agent state valu procedur find state satisfi agent integr constraint repair reset integr constraint repair code call problem longer problem iter agent cycl messag longer trigger error repair origin reset wait messag handl step simpl handl rest focu step easi step formul classic ai plan problem current state set goal state ic set ra repair action find sequenc action ra yield goal state ra complet set multipl consist state agent transit agent transit repair state wrt state evalu function classic ai plan problem confin cost function evalu state set solut ai literatur easili adopt find state wrt cost function definit state evalu agent agent state integ definit state agent optim wrt agent state iff agent state satisfi rr agent reconsid set sum ham distanc posit robot api function comput ham distanc point integ plane formal set result grid grid grid describ agent state assum grid current agent state grid grid grid grid grid grid optim state wrt satisfi grid integr reach seri action west robot locat corner sum ham distanc maxim note state optim grid grid grid optim state wrt goal state violat agent integr find sequenc repair action yield optim state easili ai plan problem major differ ai plan cost plan typic sum cost monoton function action repair action assess cost state captur state evalu find reachabl state satisfi integr constraint maxim defin state definit code call condit involv integ variabl variabl specif assum exist special packag math support number standard arithmet includ binari oper integ binari function point express grid grid math variabl specif math math variabl specif intend mean ham distanc robot sum ham rr definit code call condit math code call condit input variabl output variabl code call condit safe modulo involv code call atom access math math variabl specif input variabl output variabl variabl specif code call condit math code call condit input variabl output variabl variabl specif math math math variabl specif input variabl output variabl definit function object function specif pair hx math fv gi variabl specif form math math variabl specif input variabl output variabl function continu grid describ object function assign higher valu state robot object function specif hv math fv math math math grid grid math grid grid math grid grid math object function specif measur agent state set ground ground valu math set ground true wrt fx return object continu mean assign xn code call condit reorder evalu left appendix rr relev agent framework shown defin agent cycl appli impact agent corrupt cycl agent continu process unaffect request repair state affect request process effect natur question result paper appli agent framework show context agent agent framework framework due rao georgeff ration agent framework due wooldridg briefli discuss section kowalski sadri unifi agent architectur kowalski sadri analyz similar differ ration reactiv agent architectur unifi architectur aim captur special case agent reason captur proof procedur logic program style search engin reduc goal subgoal manner complet proof reduct procedur paper base observ case replac goal set rule problem control reason process work correctli bound resourc address result cycl govern architectur observ input environ time record resum execut proof procedur current goal propag input continu appli proof procedur total infer select atom action respect time execut action record result extens cycl account complex glanc integr constraint dynam evolv execut cycl goal reduct replac goal statement simpler goal statement form integr constraint case impact integr constraint establish made assumpt extens bdi architectur phrase kowalski sadri integr constraint mean repres condit check current necessarili satisfi shown exampl proof procedur continu execut integr constraint lead goal unsatisfi integr constraint account propag input replac current goal statement simpler account observ input integr constraint agent behavior rr avoid set static integr constraint includ knowledg base kowalski sadri static integr constraint mean impact integr establish repair procedur immedi start unifi agent architectur cycl modifi outlin observ input environ time record check input violat static integr violat start repair procedur concurr evalu resum proof procedur current goal statement lead conflict current repair atom action action unrel current turn execut continu appli proof procedur total infer select atom action respect time execut action record result interrupt process input complet repair procedur kei obstacl appli definit determin mean occur proof procedur repair procedur address wai determin atom affect repair procedur affect proof procedur intersect declar conflict notion similar paper altern solut simul advanc occur proof procedur syntact check rao georgeff bdi architectur bdi architectur base notion agent intent system excel architectur depict figur repres knowledg conjunct disjunct achiev test current procedur knowledg agent mainten set set partial instanti plan typic bdi engin cycl rr belief plan librari bdi engin goal intent action event figur bdi architectur observ world agent intern subsequ updat event gener plan trigger event match event event queue context select set match plan push select plan exist intent event select intent topmost plan execut step current step perform post event queue order extend agent cycl handl modifi step cycl event event queue safe process select plan execut safe execut select intent stack step topmost plan safe execut integr constraint hold integr constraint repair current state recov correct state modifi step bdi agent cycl step safe event event event plan definit plan trigger event match chosen event context satisfi step safe select plan order decid plan safe execut lead inconsist due repair made evalu consequ execut execut push select plan exist intent select newli modifi stack lead execut action affect plan give detail kind action affect repair scope paper good definit adapt bdi framework ignor intent stack analyz step cycl rr plan chosen step unsaf intent stack chosen step problem action perform current repair intent stack step safe select intent previou evalu consequ execut step topmost plan stack action execut potenti affect intent stack select plan select execut step intent stack chose modifi step moment event event execut cycl stop repair complet repair event event queue definit bdi cycl outlin observ world agent intern subsequ updat event queue violat integr constraint start repair procedur concurr event exist event queue event event gener plan trigger event match chosen event context select set match plan provid plan push select plan exist intent event select intent topmost plan execut step current step perform post event queue interrupt process event complet repair procedur wooldridg comput system chapter phd wooldridg formal model intend captur aspect varieti agent system model base agent finit comput agent set explicitli repres belief reason belief accord comput resourc afford belief express logic addit agent capabl commun agent belief mean belief function agent system continu execut rr interpret messag updat belief previou action messag deriv deduct closur belief deriv set send deriv set appli wooldridg defin execut model synchron asynchron agent synchron model begin end execut cycl realist asynchron execut agent act fix point time naiv extens agent cycl defin interpret messag updat belief previou action messag check belief violat agent integr violat start repair procedur concurr deriv deduct closur belief deduct closur belief integr constraint deriv set send deriv set appli interrupt process receiv messag complet repair procedur assum agent set static integr constraint satisfi state messag action messag action current repair procedur determin belief ongo repair defin adapt notion paper case wooldridg syntax relat work work error toler agent system compar work relat work area previou shown idea paper impact agent appli agent system inconsist sourc inform servic requir satisfi integr constraint ic sourc inconsist state interact sourc user place recoveri inconsist complet clear exclud rr user interact inconsist data sourc benefici research provid consist servic inconsist data sourc widespread reason inconsist studi extens context case reason multipl knowledg base notion agent cycl agent continu run process steadili execut request cycl modifi error toler process method incorpor cycl kei contribut paper import effort deal consist queri answer inform system inconsist ic bry approach make recogn answer queri deriv possibl corrupt data exploit notion local formal term minim logic data ic violat consid potenti corrupt answer establish deriv minim potenti corrupt data call inconsist answer consist comput data involv ic violat show minim logic suffic foundat queri answer definit deduct problem address context agent system access extern data sourc code call rule involv action deontic modal consist inconsist answer comput address repres differ bry approach provid algorithm evalu corrupt affect formal prove item correspond intuit notion resp chomicki provid logic consist queri answer relat inconsist ic answer queri pose violat ic obtain minim repair version method comput answer analysi properti basi queri method iter queri evalu arbitrari state return set consist answer origin queri base notion residu context semant queri optim sound approach complet ic termin comput proper condit variant oper proven termin complet class ic extend annot predic calculu logic inconsist inform unravel logic infer inconsist reason inconsist defin repair queri answer approach understand result provid gener algorithm cover class queri main differ approach consist answer evalu rewrit queri account evalu answer rewritten prove consist answer origin evalu process incom messag involv process messag yield case result ic defer complic work extern request made agent state modifi repair repair constraint interest approach basic concept diagnosi adopt discov minim set simultan reason minim logic constructivist weaken classic logic defin term natur deduct proof system depriv absurd rule rr violat constraint reason repair action elimin violat adopt repair action depend user strategi domain independ rang minim undo consist complet violat transact user interact repair process sound complet algorithm minim repair transact inconsist repair strategi similar user strategi suitabl applic applic independ librari eventu prioriti prefer repair interact user repair process briefli sketch assum suitabl environ exist environ user interact repair process easili perform impact set agent framework problem tackl solv agent work state ensur agent recov corrupt state critic agent commun agent find applic domain telecommun process control electron commerc reliabl multiag system kei issu continu recoveri properti support servic error toler awar research agent environ formal develop agent import aspect endeavor build ration make good necessarili situat turn agent ahead action make underli assumpt integr constraint correct imagin integr constraint withdraw modif integr constraint remedi situat agent choic agent state integr part agent entiti access plausibl agent design inform violat integr decid chang integr constraint notion auxiliari technic concept formal intuit action unrel error peculiar make nontrivi task result section show techniqu adapt type agent architectur work agent commun relat shoham explicit program languag base object orient base concept agent state shoham section state complet aop orient system compon restrict formal languag describ mental interpret agent program languag primit command request rr impact agent express shoham aop framework result paper appli shoham aop framework hindrik al develop deontic base agent program framework agent mental state consist set goal set belief agent program framework consist quadrupl transit function effect basic initi set initi set set rule form head guard head complex formula describ goal syntax goal support hindrik goal elementari includ sequenti composit disjunct conjunct goal guard logic bodi structur head hindrik al express impact vice agent cycl similar result error toler appli agent cycl appli paper impact agent cycl case agent make framework initi framework rosenschein agent act action determin rule form true state agent action cycl similar state chang determin base rule involv notion directli rule involv unaffect atom process request repair mechan conduct repair agent state appli irma system bratman al agent gener action base agent intent plan evalu determin consist optim respect achiev intent cycl similar plan involv atom softwar agent provid power paradigm mobil applic softwar agent system build top legaci softwar principl support automat coupl simpl complex action wide varieti applic danger assum agent prototyp declar languag impact free error prolog program year histori program shown bug code account impact base agent agent system build top legaci bug reason agent develop written rule account state agent agent full control state true legaci applic agent vehicl access legaci applic state legaci code top agent built unexpect agent state modest step address difficult problem time agent algorithm good featur incorpor method agent recov state rr agent continu process request long request unaffect ongo repair work contribut endeavor build ration agent shown method appli agent bdi model import aspect reason behavior bdi model discuss demonstr wooldridg ration remain interest issu model extens observ build top classic agent state integr constraint model set classic fact violat integr constraint conclud avoid method paraconsist logic suitabl belief oper contribut paper admittedli panacea problem involv bug agent program handl case agent dont statu set due violat integr constraint violat occur parti manipul agent state agent updat agent rule adequ deal ic violat scenario repres small space error agent program form rich avenu futur research result paper relax futur work wai action atom view action statu atom determin suspici code call atom basi syntact analysi agent program similar paper due interplai semant compon feasibl statu set includ deontic action closur integr provid refin approach studi substanti complex acknowledg review construct comment improv paper work support part austrian scienc fund project armi research lab contract number aro grant grant refer vianu foundat addison chomicki consist queri answer inconsist proc pp acm kifer applic annot predic calculu queri inconsist proc juli subrahmanian platform collabor agent ieee intellig subrahmanian combin multipl knowledg base consist order theori comput rr blair subrahmanian paraconsist logic program theoret comput pollack plan practic reason comput bry queri answer inform system integr constraint integr intern control inform vol confid inform ifip wg work confer integr control inform system chapman decemb cell queri inconsist algorithm implement proc juli jen part electr transport manag ieee williamson middl agent internet proc intern joint confer artifici pp intent stanc mit eiter subrahmanian heterogen activ algorithm complex artifici pick heterogen activ semant artifici roger heterogen activ polynomi implement agent artifici extens framework repair constraint violat integr intern control inform vol confid inform ifip wg work confer control inform pp chapman decemb de van der meyer formal semant abstract agent program languag intern workshop agent languag volum pp kowalski sadri unifi agent architectur combin ration reactiv proc intern workshop logic volum pp san kowalski sadri agent architectur unifi ration reactiv imperi colleg technic kowalski sadri logic program system annal mathemat artifici rr nilsson principl artifici intellig morgan rao georgeff model ration agent fike proc knowledg reason pp san morgan kaufmann publish rosenschein formal theori knowledg ai robot gener shoham agent orient program artifici subrahmanian paraconsist disjunct deduct theoret comput ross heterogen agent theori implement mit sycara zeng integr inform gather support european confer artifici intellig intellig agent telecommun jen agent applic market pp wooldridg logic model comput system phd depart octob wooldridg reason ration agent mit reason statu set appendix succinct form definit concept statu set reader refer inform definit statu set set ground action statu atom valu type domain softwar packag oper op denot set sg definit app agent program agent state app true predic true iff ground ng definit statu set deontic action ground action case impli impli impli statu set denot smallest set close action close rr definit statu agent program agent state statu set feasibl statu set app ground action impli impli action state result execut execut strategi action satisfi integr ic definit ration statu statu set statu set exist satisfi condit feasibl statu set statu set ration statu feasibl statu set ground definit statu agent agent statu set negat action statu atom occur reason statu set iff ration statu set reduct wrt denot red program obtain ground instanc rule remov rule bodi remov neg liter remain rule reason statu set wrt reason statu set program red respect safeti variabl root involv object variabl involv root variabl refer object definit code call code call safe iff ground code call condit safe iff exist permut comparison op constant variabl belong rv st occur constant variabl rv root variabl rr code call atom form cc cc root variabl occur cc belong rv root rv code call reorder code call atom occur evalu atom left assum root variabl increment bound object definit modulo code call set root variabl safe modulo iff assign object variabl case safe 
AI/1856 2 trend evolutionari robot lisa comput scienc program colleg pa usa kumar depart math comput scienc colleg pa usa abstract review evolutionari techniqu automat design adapt robot focu method neural network test actual physic robot chapter examin role simul domain knowledg evolutionari process conclud predict futur direct robot soft comput intellig robot edit lc jain introduct robot adapt collect basic abil brought bear tackl varieti task wide rang environ fundament abil includ navig goal obstacl object object manipul desir level adapt realiz robot primarili success deploi constrain environ perform determinist task result robot compet gener situat main advantag strict constraint robot domain enabl predict model impli set sensor read accur translat current state world result action advanc execut robot plan behavior work simplifi cost industri robot system program excit potenti applic involv complex dynam environ typic attempt planet expect burden human design control softwar move advanc applic hope uncertainti variabl occur physic sensor actuat lessen hardwar technolog allevi program problem experi shown higher resolut sensor introduc expect evolutionari comput avenu overcom bottleneck human engin robot design process evolutionari approach design robot eventu design hand fundament idea approach maintain popul robot control architectur initi popul typic collect randomli configur architectur architectur evalu object fit measur robot perform architectur offspr produc gener popul number fit popul success architectur creat human engin develop evolutionari actual design robot system automat gener accur fit measur potenti architectur control actual robot oper physic world slow process action requir time execut fit measur typic involv hundr action process singl gener requir hour time hundr gener achiev reason run evolutionari process requir dai techniqu emploi speed evolutionari simul faster fit evalu seed initi popul domain knowledg number gener requir central question adopt evolutionari comput approach type randomli initi determin fit repeat select parent recombin mutat parent creat determin fit individu good figur gener evolutionari comput algorithm robot control architectur number code machin code paramet set system rule entir strategi innov combin evolutionari comput artifici neural network neural network evolutionari process oper low place minim constraint solut architectur design bia preval neural network toler local learn start point discov global evolutionari process remaind chapter organ section background evolutionari comput neural network section specif research project combin evolutionari comput neural network control robot examin section current debat role simul review section wai domain knowledg incorpor evolutionari robot system discuss final section specul futur trend robot explor background evolutionari comput major current implement evolutionari comput descend branch genet evolutionari evolut strategi genet algorithm approach develop basi evolutionari robot applic relat techniqu genet algorithm classifi system genet program evolutionari comput method attempt mimic process natur evolut gener structur evolutionari comput algorithm shown figur depend mutat select implement probabilist process rel fit individu determin select probabl fitter individu particip produc gener recombin mean mix genet materi parent produc offspr mutat creat random alter genet materi offspr effect oper select fill popul copi individu initi popul mutat induc random walk search space select mutat creat search select recombin process converg solut oper form global search method artifici neural network common artifici neural architectur layer feedforward network unit layer link weight connect unit layer amount activ appli unit initi repres input network activ flow connect higher weight mediat amount activ pass success unit final pattern activ present layer consid output produc learn algorithm repeatedli appli adjust weight network enabl learn associ arbitrari pair input output pattern input pattern interpret repres sensori inform output pattern network control robot result train network learn recod incom sensori pattern pattern intermedi layer output action produc intermedi layer term hidden direct access environ drawback standard feedforward architectur difficult repres time effici manner control robot time inform crucial common method accommod time repres time space entail extend input layer act buffer past approach deal time implicitli explicitli recurr architectur recurr call elman unit hidden layer weight connect unit includ network memori intern recod past sensori input radic recurr architectur abandon layer topolog unit connect arbitrari wai section exampl arbitrari network architectur emploi control robot combin evolutionari comput artifici neural network appli evolutionari comput neural network robot main method method fix topolog network architectur evolutionari comput determin weight unlik gradient descent procedur find set evolutionari comput avoid stuck local minima method evolutionari comput determin aspect network topolog weight set separ learn algorithm determin evolutionari comput section exampl evolv weight evolv develop robot control exampl evolv neural network robot control conduct evolut physic robot time consum result applic attempt fairli modest scope reason robot small task environ set desktop robot tether comput data collect tether electr outlet power popular platform conduct evolutionari experi khepera robot khepera circular shape miniatur height weight dc motor power wheel standard sensori apparatu consist proxim sensor applic khepera batteri recharg trash collect small call locat visual target miniatur call light seek avoid drive task involv full size car call part ongo project mellon batteri recharg environ empti rectangular arena grai floor light emitt light circular patch floor paint black repres batteri recharg area robot happen pass black area virtual batteri instantli recharg khepera robot equip addit sensor task light gather ambient point floor virtual batteri level sensor goal evolutionari process determin set weight fix recurr network twelv input unit hidden output unit motor process run continu ten dai gener approxim quarter hour individu set weight evalu simpl fit function consist maxim speed avoid wall robot recharg fit defin notic explicit reward recharg fact term direct fit detriment spend time recharg area implicit benefit recharg fulli charg batteri robot move time individu upper limit time step evalu recharg robot gain time accumul fit analysi result behavior reveal robot exhibit strategi depend batteri level batteri level drop full robot began execut trajectori led recharg area move full speed slightli bend trajectori avoid recharg area gener fit function evolv robot control locat batteri recharg area time home maneuv batteri level low extend gener floreano mondada suggest interest redo experi elimin fit function simpli select individu live longer make artifici evolut process similar natur evolut trash collect order pickup khepera robot equip gripper modul environ rectangular arena piec randomli distribut trash cardboard robot task find pick carri boundari drop nolfi compar determin modular task feedforward network motor output depend competit separ modul number modul fix output interact modul determin adapt weight fit measur primarili depend target successfulli arena bootstrap fit measur includ compon reward robot abil simpli pickup target prove difficult robot learn react target carri target allevi train experi manipul make type occurr frequent train control evalu simul spite evolut run requir approxim ten hour train result control system download physic robot test real environ succe abil distinguish wall target crucial import error made robot reveal simpl attempt grasp wall target target emerg modular architectur creat modul specif tune make distinct sensori situat capac produc sharp switch strategi sensori differ kei complex behavior locat visual target visual task includ locat larg static locat smaller static track move locat triangl target presenc compet rectangl target experi special piec robot equip develop call robot robot suspend frame translat movement direct robot diamet rotat equip camera point mirror inclin bumper sensor evolutionari algorithm search network architectur visual morpholog network fix number input node output node number hidden node number type connect excitatori variabl connect recurr layer feed raw camera imag method sampl imag evolv network achiev design set recept field imag visual task involv find light color target dark color environ fit measur function distanc robot desir closer robot target higher fit architectur visual robot trial start posit orient final fit worst score trial termin trial previou trial improv evalu time start evolutionari process initi random popul prove slow select member initi random popul displai behavior clone creat popul work converg popul random popul term speci adapt genet algorithm start evolut run develop good solut gener gener half hour robot evolv successfulli locat larg static target final popul experi start point experi locat smaller static target gener popul adapt successfulli harder task popul recycl start point harder task track move target work increment fashion gradual build complex behavior morpholog robot develop control simplifi task solut seek avoid light modifi toi car approxim wide long equip light sensor bumper sensor rectangular arena light corner task constantli avoid respond appropri light goal goal seek light maximum light read obtain accomplish goal automat switch neg indic avoid light minimum light read obtain goal vari period manner seek immedi avoid local global method reinforc learn compar train special form evolutionari algorithm topolog network fix elman network input sensor hidden output unit motor set aim method find set weight adapt process conduct simul architectur test actual robot statist reveal quantit differ learn method algorithm evolutionari algorithm origin task evolutionari algorithm algorithm task difficulti remov explicit goal period remov reinforc feedback interest qualit differ behavior produc method local sensit environ explicit goal develop uniqu strategi tune goal fact explicit goal train network creat unit hidden layer evolutionari algorithm tend develop singl strategi applic goal evolutionari algorithm abil find good strategi robust variat respect strength weak method suggest hybrid effect method evolutionari algorithm global sampl entir space altern solut backpropag local search neighborhood straightforward form hybrid evolutionari algorithm find good start point weight space fine tune global evolutionari method determin good gross solut local learn method adjust current condit drive autonom land vehicl oper wide varieti domain includ bike neighborhood divid highwai control determin steer angl video imag camera mount front car unlik previou task clear answer situat determin human driver number network architectur explor solv drive success feedforward topolog input retina video small hidden gaussian steer angl thirti output unit rang determin weight network control baluja set discov evolutionari comput develop control task network describ maximum connect network evolv topolog test select connect enabl connect weight determin maxim network experi input retina fulli connect unit hidden layer fulli connect thirti unit output strictli feedforward order reduc search baluja creat evolutionari method call increment learn algorithm requir individu popul repres binari alphabet weight repres binari number restrict rang goal produc probabl vector consid prototyp highli fit individu popul sampl singl probabl vector entir popul stochast produc test base test result valu adjust push probabl vector individu worst individu mutat oper updat probabl vector shift small posit neg amount unlik genet recombin oper test network drive train set imag correct steer angl collect save exampl randomli select network evalu network obtain lowest error design gener network highest error worst final network evolv method half connect maxim network compar maxim network train network show effici accur control gain complet entir run requir hour run requir minut process time summari exampl batteri recharg floreano mondada demonstr complex behavior obtain overli explicit fit function suggest emploi human engin measur ultim measur gener fittest contrast call implicit fit cliff argu complex behavior obtain set design increment harder fit test relat visual locat show begin evolut run converg popul random faster explor produc import aspect cliff experi visual morpholog robot evolv control human sensori apparatu robot sensori human process visual imag translat robot evolutionari process oper sensori effici solut emerg similar drawn nolfi respect modular comparison number network architectur trash collect nolfi found provid network modular option extrem benefici differ human robot sensori human subdivid task translat robot evolutionari process break task lead simpler solut light seek avoid demonstr evolutionari algorithm robust method determin good gross solut robot control problem suggest solut improv fine tune addit ing local learn method hybrid kind robust larg sensit subtl featur final baluja evolutionari comput method call emploi creat control drive topolog weight network determin evolutionari process result demonstr automat design control control question studi larg role human design plai shape robot behavior fit model evolut true natur engin influenc evolutionari process relat question emerg studi aspect robot control system manipul evolutionari robot featur access evolutionari process success adapt control simul hot debat peopl understand reproduc intellig state simul power tool draw sound theori approach test real simul control ai complet wast time case simul inevit issu result simul accord worth simul suggest simpler case shown made accur attract qualiti speed data collect made easier robot situat real world build encompass make robot distinct artifici intellig project deal hardwar limit physic act system difficult hardwar interact issu simul debat term refer simul physic counterpart model fact simul implement real world depend provid inform environ danger simul simplifi learn problem make environ robot clean predict result obtain simul transfer noisi real world term difficulti simul present attract primarili speed evolutionari experi dai physic robot requir hour simul robot abil obtain result quickli facilit explor robot control idea simul result easili reproduc lessen correspond effort made simul close step realiti specif suggest made base simul carefulli collect empir data real physic add simul sensori read actuat calibr simul test real physic robot calibr test provid interest contradictori result neural network robot control adapt simul perform test real world test simul suspect improv physic robot movement sensor read noisi wai simul physic robot experi noisi simul experi systemat benefici learn experi conduct determin amount simul affect correspond behavior evolv simul test simul versu test real world level observ doubl observ result reveal network evolv environ noisi real world behav realiti network evolv environ real world behav realiti correspond simul realiti maxim level simul close match realiti case addit benefici case contradictori result resolv experi observ drastic chang realiz smaller increment observ issu amount open question relat simul adapt robot evolutionari option evalu control physic simul hybrid physic robot desir time consum simul robot lead correspond problem hybrid begin evolutionari process simul robot quickli develop high perform popul assum simul develop close observ actual provid good start point slower evolutionari process physic robot simul robot quickli prune set experi eventu conduct physic robot hybrid approach evolutionari robot offer speed accuraci domain knowledg role domain knowledg evolutionari robot view continuum robot complet domain knowledg requir abil extrem robot domain knowledg rasa system recent workshop robot research present incorpor substanti amount domain knowledg present argu construct success robot learn limit portion system design knowledg sketchi engin solut learn scratch ineffici problem reason complex middl ground continuum approach treat knowledg acquisit cooper effort human engin robot grefenstett develop evolutionari method stand strategi acquisit method empir learn samuel evolutionari process develop entir behavior defin set rule rule express languag make easi incorpor exist knowledg make easi understand result learn initi popul behavior random consist varieti rule set includ human gener automat gener variant seed initi hope search space usefulli constrain lead faster search time back section techniqu constrain search creation virtual sensor base rule sixteen raw sonar creat virtual combin raw valu begin amount domain knowledg practic disadvantag perspect learn problem make harder nolfi argu design view robot task observ point view distal robot solv task term sensorimotor system proxim perspect simpl map distal perspect proxim perspect trash collect task section distal descript task explor environ avoid wall recogn trash approach pickup pickup trash move wall avoid trash trash wall recal khepera robot experi access sensor human vision distal simpl matter distinguish wall robot sensor proxim object distinguish small number close posit modular task distal subtask wall avoid viabl trash wall local view argu artifici neural network offer mean investig robot control task demand design primari forc shape system develop design import role plai evolutionari process determin aspect neural network oper input output robot physic characterist manipul robot environ adapt process set motion success control system method understand underli control principl evolutionari comput neural network invert classic order problem solv understand close guid search algorithm evolut neural network control solut task emerg simpli product design understand domain futur trend robot han moravec todai robot simpl global evolutionari predict decad robot improv level year level mammal crucial impedi model adapt behavior lack model size speed comput emploi evolutionari techniqu develop neural network base control progress made recent partli due continu exponenti comput resourc moravec point amount comput power dollar decad begin centuri declin cost comput trend case moravec predict comput power requir humanlik robot person comput year work humanlik robot begun mit cog project approach build humanoid robot develop act real world human develop act develop plan led incorpor arm reflex arm adapt arm follow smooth motion ey foveat behavior coordin movement reach ey head exhibit saccad motion gaze control cog perform todai convent expect approach eventu pai model incorpor cog base biolog model suggest combin artifici intellig evolutionari comput repres innov research direct lead develop solut complex problem comput power increasingli attract experi evolutionari method robot onboard comput mobil system power run experi simul continu plai role time feasibl incorpor robot school colleg robot preval expect rapid innov futur refer schwefel evolutionari comment histori current state ieee transact evolutionari brook stein build brain bodi autonom evolut artifici neural network base autonom land vehicl control ieee transact clark associ chang mit husband explor evolutionari robot adapt editor artifici real mit press dorigo behavior analysi methodolog behavior engin ieee transact dorigo preci robot experi behavior engin adapt elman find structur time cognit ferrel orient behavior regist topograph map anim intern confer simul adapt floreano mondada automat creation autonom genet evolut driven robot anim intern confer simul adapt page mit press floreano mondada evolut home navig real mobil robot ieee transact goldberg genet algorithm search machin learn publish gomez miikkulainen increment evolut complex gener behavior adapt grefenstett evolut strategi environ adapt schultz learn sequenti rule simul model competit machin grefenstett schultz evolutionari approach learn robot naval research laboratori technic report harvei evolutionari robot case hill crawl tournament select artifici life page addison cliff issu evolutionari robot anim intern confer simul adapt page mit press horswil softwar architectur hardwar agent journal theoret artifici learn autonom summari workshop knowledg engin holland adapt natur artifici system michigan ann miller artifici path artifici brain harvei realiti simul evolutionari robot advanc artifici european confer artifici page pomerleau life fast evolut adapt vehicl control system ai koza genet program comput mean natur select mit bradi softwar hardwar architectur advanc mobil robot manufactur journal theoret artifici williamson visual guid point humanoid robot anim intern confer simul adapt increment approach develop intellig neural network control robot ieee transact robot introduct comput scienc ninth florida artifici intellig research page blank emerg control plan autonom vehicl fifteenth annual meet cognit scienc page lawrenc erlbaum associ mitchel introduct genet algorithm mit mobil robot tool investig control algorithm intern symposium robot vision technolog comput page st martin nolfi emerg modular develop control system mobil robot adapt mondada evolv autonom approach evolutionari robot brook artifici live fourth intern workshop synthesi simul live page mit press banzhaf method evolv behavior control miniatur robot real time genet program adapt pomerleau neural network percept mobil robot guidanc pearc genet algorithm learn reactiv control paramet autonom robot navig adapt reynold evolut corridor behavior noisi world anim intern confer simul adapt page mit press william learn intern error propag mcclelland parallel distribut volum page mit robot make harder anim intern confer simul adapt page mit press van de veld learn robot robot autonom 
ML/2836 6 learn case resolv conflict improv group behavior thoma hayn sandip sen depart mathemat comput scienc south colleg avenu abstract group agent fix behavior rule limit perform effici adapt flexibl kei compon intellig behavior agent group improv perform domain prior problem solv experi motiv individu learn group member context group behavior framework individu group member learn case improv model group member testb problem distribut ai literatur show simultan learn group member lead improv group perform effici agent group static behavior rule introduct agent defin ration face choic set maxim expect util action implicit definit assumpt prefer agent action base util result action problem multiag system action agent conflict agent agent model behavior incorpor expect util calcul group optim action individu agent optim action entir group action evalu respect individu group benefit agent evalu util action individu group group level calcul requir inform individu level calcul yield desir result agent group util calcul individu perspect requir reason action group member individu group model behavior group model reason action analysi hold irrespect agent agent briefli evalu typic problem face agent model group member agent start approxim model group member start default assumpt modifi model base experi agent interact unforeseen dynam model group maintain individu problem model agent base passiv observ discrep expect actual observ agent lead infer model inaccur agent perceiv view environ observ agent correctli infer motiv action action interv time agent act asynchron agent commun improp underli prevent agent develop share common bodi knowledg commun intent negoti avoid conflict situat prove time consum impract domain problem combin confound individu attempt predict behavior member group investig method agent improv model member group evolv model agent determin local action goal show gener behavior rule effect achiev local goal absenc ineffect share resourc group agent learn modifi behavior achiev goal presenc agent assumpt work agent provid default set behavior rule repeat interact group member agent modifi behavior rule agent motiv achiev local goal cogniz global agent agent percept agent commun agent act adapt concurr agent interact agent fairli infrequ environ static set behavior rule suffici effect fulfil local goal similar argument made cooper agent group environ understood effect group behavior design larg number practic interest agent interact agent unknown composit agent interact foreseen adapt learn kei mechan agent modifi behavior maintain viabl perform profil scenario number research recent start investig learn approach target multiag system elimin commun group learn actual outcom action agent consist expect outcom base model agent agent found case model agent default behavior rule inadequ case base reason effect deal situat research cbr multiag system work learn case multiag system learn framework agent learn case complement behavior rule agent find interact agent behavior situat learn except behavior rule follow behavior rule learn case guid act agent dynam evolv behavior suit group narrat typic multiag situat case learn effect adapt local group learn commun interact adam cat receiv shot food shot adam decid administ shot wake work discov react sound food dish alarm clock cat learn wake adam food except routin learn cat expect alarm clock morn met solut place learn mechan top individu level mechan adapt individu greedi strategi local global goal learn algorithm util except default describ behavior agent except form case librari agent reason cbr modifi inaccur individu model approxim group model form model complimentari method postprocess output agent model postprocess place potenti interact agent agent ultim goal select action method usabl rest paper organ section present reason defin learn multiag system section discuss domain suitabl testb research cbl section present result appli cbl multiag domain section discuss implic find testb domain multiag system gener learn intellig defin delic balanc search relev applic modif knowledg applic fail knowledg repres collect rule human tend default rule cover major situat emploi evid present show except hold difficulti detect except vari domain reason model definit reason process inform retriev modif solut problem case typic state domain set action lead state desir state action modif case action case librari collect case cbr algorithm modul determin case match current state retriev case similar current state retriev case librari set action relev case adapt fit current situat cardi defin learn machin learn techniqu extend learn ibl algorithm retriev nearest instanc instanc thought perform suggest action case adapt retriev instanc direct match current state adapt place view case gener set instanc viewpoint context multiag defin learn system agent extend default rule respond except rule adapt li translat gener case specif instanc case system agent preferenti order action singl agent state repres multiag repres environ agent expect action agent present formal cbl system tailor multiag system case behavior rule agent thought function map state applic action set agent prefer order case agent learn modifi prefer case fire time agent perform case posit neg posit case inform agent reorder set action neg case reorder action delet action set case system present paper neg elimin prefer action suggest behavior rule agent learn agent learn case expect fail behavior rule case predict state applic action agent expect state agent end case learn agent case action consid time agent state multiag expect case learn primarili unexpect interact agent case gener elimin irrelev featur state agent influenc state expect behavior includ index creat case model agent behavior rule agent learn except default check except modifi time system agent modifi behavior check action case behavior rule updat model concret problem dai literatur illustr approach cbl multiag system domain wide distribut ai research testb investig cooper conflict resolut goal predat agent captur prei agent block orthogon movement game typic plai grid korf claim commun captur simpl greedi strategi lead captur hayn al show greedi strategi captur prei strategi perform poorli prei behavior pick direct move move behavior strategi predat distanc manhattan distanc max norm md metric sum differ coordin agent mn metric maximum differ coordin agent algorithm examin metric set move orthogon direct stai select move minim distanc metric ti randomli broken md strategi success mn captur linear success low spite fact block forward motion reduc prei answer li fact md metric algorithm suscept deadlock situat situat shown figur greedi natur famili algorithm ensur situat similar figur predat yield predat predat predat mn metric algorithm perform linear prei advanc analysi explain lack captur prei question find agent manag conflict answer found wai human manag conflict case simplest predat predat determin move agent move west figur scenario md metric base predat block prei predat manag block predat move captur predat move captur posit predat move captur posit predat yield predat prei captur conflict predat predat learn move north prefer direct motiv learn case research examin multiag learn potenti conflict default rule emploi predat move closer overrid case present case move consid suggest default rule case case learn move md algorithm suffic predat agent equidist prei prei effect move slower predat grid world prei move predat order move predat eventu catch predat close bunch orthogon content desir cell start plai predat agent vie greedi natur algorithm overridden simpli order movement predat fastest captur prei static order effect situat requir commun combat engag fighter aircraft order rule suggest fighter move fighter situat depict figur fighter fire hit fighter fighter move north fighter clear fire lane order rule move manner default rule fighter move fighter except fighter front fighter rule modifi agent front move add situat figur handl figur conflict fire lane fighter plane fighter block fire fighter fighter fighter fighter fighter fighter decid move cell north fighter problem discuss md algorithm dynam learn mechan model action agent potenti conflict agent follow default behavior conflict occur agent learn agent act specif situat agent learn emploi default rule situat consid action specif situat encount form librari conflict avoid agent learn begin model action group agent start rough model develop refin individu model agent group enhanc behavior rule order facilit effici provid agent set default decid enhanc basic md algorithm provid challeng learn algorithm improv human agent plai reason emploi simpl greedi behavior rule refin deriv fact human face equal attract human spend extra comput effort break tie introduc human bob bob make alex alex repli similar situat dinner bob decid king denni trick colleg decid choic predict react restaur denni step situat shown figur predat cell denot cell prei manhattan distanc metric sum distanc move prei similar sum figur cell mark equal distant manhattan distanc metric prei predat chosen sum move prei sum chosen ahead potenti conflict predat call ahead predat chose refin ahead yield equal distanc scenario shown figur predat determin cell content agent predat mind contend cell wast move bounc predat conflict predat pick bob time alex fight open town par bob face break tie alex believ alex make bob end join coupl night conduc decid model alex behavior alex student eaten denni alex chang alex bob decid notic bob account chang case conflict ultim goal final enhanc restructur order action stai action consid benefit predat captur attack prei forward move predat learn move prei predat follow prei move move predat learn avoid deadlock situat case index ideal case domain store entir world case inform predat move problem number case agent act independ agent store case librari size problem effect predat repres cell directli figur simpl case window predat figur case predat stai denot area prei occupi conflict move shown figur drawback approach agent present case part case predat locat impact desir move east predat figur problem predat north predat predat prei move differ scenario present number case larg entir world narrow case window suffer point case window present case repres actual configur domain model accept case window map physic situat case gener appli multipl issu find relev gener case case window figur encompass potenti conflict agent move direct limit case window simpli repres potenti conflict occur agent select move base default rule learn util case window shown figur case window figur symmetr direct agent leav cell case window figur appli agent remain stationari symmetri agent engag cbr case neg agent posit case agent situat crucial piec inform decid local action agent agent model store orient prei posit respect desir direct movement agent store prei li agent line advanc left line agent combin behavior rule learn case action figur predat learn wrong case predat action move prei predat bounc back prei store incorrect fact stai move east predat push predat predat learn incorrect case action stai configur shown select action agent prepar order action default rule md distanc metric addit iter rank check neg case move index agent determin action movement stai discuss determin case librari access examin content cell case content conflict content sum form uniqu integ index base number system reflect rang content action neg case chosen move turn preliminari case window shown figur learn captur prei result found predat learn hinder effici captur prei erron scenario shown figur problem li predat interpret scenario move locat prei fail learn situat involv prei predat push predat predat learn yield predat face situat shown figur case learn stationari case window shown figur case inform predat current action stai action case fire predat action move prei ps predat learn case case remain action exact conflict predat case appli case learn predat situat depict figur action figur situat learn current case window order captur decid merg case window figur case window case window shown figur agent store case potenti conflict resourc desir resourc possess learn case shown figur figur case fire predat action enhanc move north figur case predat move predat case stai figur case window note show predat learn case setup result initi configur consist prei center grid predat random posit agent action simultan environ updat agent action base updat environ state agent move locat prior posit push predat decid move prei move effect make predat travel faster prei grid orthogon move agent posit agent predat possess explicit commun predat commun resolv conflict negoti captur strategi case window emploi depict figur initi interest abil predat behavior rule effect captur prei test behavior md basic md md modifi enhanc discuss section util case base learn train random simul result appli strategi test case shown tabl enhanc behavior rule addit neg case lead captur simul conduct set experi prei linear algorithm behavior rule test predat behavior strategi algorithm train prei train prei shown section linear prei typic degrad prei present result train linear prei result linear prei present tabl algorithm captur av number step md tabl number captur test averag number step captur prei algorithm captur av number step md tabl number captur test averag number step captur linear prei denot test train linear prei prei order effect addit cbl multiag system instrument effect behavior rule room result linear prei captur rate run result agent conflict situat major time spent captur linear prei spent chase block interest conflict situat occur shown learn effect appli multiag system difficult problem group dai literatur shown significantli improv perform agent group util fix behavior rule suggest interest avenu futur research critic aspect agent group plan investig chang agent model potenti problem algorithm agent learn model group agent group refin model group interact learn model agent construct invalid model situat environ select except agent model note simpli forget learn earlier return cat present situat group learn occur daylight save time effect time alarm clock set push back hour inform chang environ adam model cat wake morn wake adam adam bed alarm sound week wake routin wait time wake adam experi order agent significantli improv perform learn essenti wide arrai situat agent experi creat interact scenario learn agent group forget order improv perform present essenti incorpor structur mechan delet overwrit case recogn ineffect import multiag system multipl agent concurr adapt behavior agent model agent bound outdat person knew person modifi learn store inform agent learn expect behavior agent work develop explod search space refer david denni marc albert learn algorithm machin januari optim cooper knowledg sourc technic report boe ai boe comput august clair cardi tree improv learn tenth intern confer machin page morgan kaufmann andrew richard prepar knowledg david aha work note aaai symposium adapt novemb aaai le nicola randal john repres organiz knowledg dai system le gasser michael distribut artifici volum research note artifici page piotr gmytrasiewicz edmund oper formal model victor intern confer page san mit press andrew gold paul rosenbloom improv system reason ninth nation confer artifici page joseph halpern yoram mose knowledg common knowledg distribut environ acm confer principl distribut thoma hayn sandip sen evolv behavior strategi predat prei gerhard sandip adapt learn multiag lectur note artifici intellig springer winter thoma roger sandip dale strongli type genet program evolv cooper strategi larri sixth intern confer genet page san morgan kaufmann reason belief base common knowledg observ action victor intern confer page san mit press janet editor workshop reason morgan janet kolodn reason morgan kaufmann richard korf simpl solut pursuit game work paper intern workshop distribut artifici page februari victor lesser multiag emerg ai acm comput septemb ran levi jeffrei rosenschein game theoret approach pursuit problem work paper intern workshop distribut artifici page februari osawa metalevel coordin strategi reactiv cooper plan victor intern confer page san mit press nagendra victor susan lander reason retriev distribut case base journal visual commun imag special issu digit umass cs technic report stuart russel peter norvig artifici modern approach prentic sandip editor work note workshop adapt learn multiag montreal august sandip john hale learn coordin share inform nation confer artifici page john steven salzberg combin genet algorithm memori base reason larri sixth intern confer genet page san morgan kaufmann munindar singh effect agent control strategi perform dai pursuit problem work paper intern workshop distribut artifici octob stephen research paradigm ai spring larri stephen matthia agent organ effector dai system perform work paper intern workshop distribut artifici septemb larri stephen matthia effect agent control strategi perform dai pursuit problem distribut ai octob katia sycara plan approach darpa plan page decemb katia sycara resolv adversari approach integr analyt method phd school inform comput scienc georgia institut jose vidal edmund agent model limit ration victor intern confer page san mit press ian watson review knowledg engin 
Agents/2142 1 revisit ration agent intent lawrenc cavedon lin anand rao elizabeth sonenberg comput scienc australian ai la comput scienc abstract formal framework specif autonom agent commonli base logic intent belief desir properti logic intent logic belief address problem exist logic intent belief shift make world framework invalid problemat properti intent constraint algebra structur show desir properti obtain world framework semant prove extrem power flexibl tool logic specif ration agent behaviour introduct logic desir intent recent receiv attent ai literatur design autonom intellig agent import logic framework abstract specif behaviour agent logic tend modal build epistem logic knowledg belief base world approach inherit problem logic omnisci problem agent model logic treat worst exhibit properti ideal agent accept logic intent displai logic accept logic belief intend oe necessarili lead intend oe intend oe agent achiev wai conjunct intent adopt treat inher problem intent treat updat problem inher problem intent respect adopt intent strategi ensur unwant intent remov modal logic gener approach elimin unwant consequ logic omnisci agent forc respect separ notion proposit object set world true introduc author anonym comment work partli support cooper research centr intellig australia notion imposs world world necessarili exhibit simplest logic behaviour approach agent proposit oe logic proposit problem world object reason model realiti evalu modal world potenti epistem state agent model view unreason ascrib state agent requir exhibit behaviour appli world approach logic belief intent focuss primarili properti intent distinguish modal properti relat attitud argu set number properti desir logic show properti highlight problem altern model world extrem flexibl framework show algebra condit easili model theoret semant obtain properti desir logic algebra condit set captur properti belief intent captur relationship attitud addit featur formal belief intent exhibit belief made requir forc similar properti intent desiderata logic intent belief number author axiom infer rule combin logic intent belief discuss theorem logic entail properti broadli classifi depend interact depend underli logic model object intent belief main properti agent intent consist belief inconsist agent belief incomplet respect intent incomplet similarli agent intent incomplet respect belief incomplet bratman refer properti collect asymmetri thesi properti receiv great deal attent literatur consequenti closur problem properti state agent intent close respect belief implic ration agent intend believ impli forc intend pain properti properti state undesir properti state formula explicitli desir ignor tempor aspect logic issu address futur work oe consist oe incomplet oe incomplet oe oe consequenti closur set properti usefulli model human model comput agent stronger version axiom oe state agent intend believ achiev agent note impli vice versa categori properti concern belief intent interact focu properti intent reason requir intent oe agent hold intent inconsist intent close consequ oe usual oe oe possibl reject author note properti oe properti oe desir respect disjunct enforc properti oe motiv avoid intent arbitrari irrelev proposit properti oe case aim avoid choic lose contextu inform prefer logic remain deal disjunct singl entiti reject properti intent disjunct properti oe intend bui bottl red wine white intend bui bottl white reason conclud intend bui bottl red wine summar categori properti intent note axiom accept talk achiev agent intent achiev oe agent achiev oe current framework abstract discuss issu section framework gener properti requir belief overlap requir disjoint intuit arent complet clear case feel strongli reject note necessarili follow disjunct closur axiom world framework desir provid systemat investig properti requir logic intent topic futur research intent consist oe intent consist ii oe conjunct composit oe conjunct decomposit oe disjunct decomposit oe disjunct composit oe disjunct closur oe oe gener intent closur oe infer intent addit properti import logic theori develop provid complex intent simpler directli oe longer hold address issu introduc notion intent end achiev end intent achiev intent achiev oe intent achiev plan achiev achiev oe achiev intent reduc search plan achiev process explor section logic intent belief world semant build idea cresswel develop world framework modal logic semant fine grain notion mean avoid logic properti retain explicit regular framework set logic knowledg belief show exist approach model explicit implicit belief awar van der hoek view instanc world semant kei idea approach add set world usual kripk frame defin truth falsiti formula involv modal oper term truth valuat world result semant full composit structur evalu modal formula model set normal set binari relat valuat satisfi note kripk model simpli model empti iff iff iff world truth falsiti formula note truth oe normal world depend truth oe normal world involv evalu modal formula oe world truth formula oe subformula gener affect truth oe valid model defin term valuat normal oe valid model oe valid valid model proposit tautolog necessit fail valid oe impli valid oe logic properti necessarili present world approach essenti background logic classic proposit logic background logic extend requir constraint admiss valuat condit binari relat world obtain condit involv requir oe belief intent extend framework world modal oper belief intent replac binari access relat binari relat relat tupl set normal set binari relat valuat modal constraint form band properti discuss belief intent section illustr world semant captur properti properti direct inspect definit model hold provid wi wi write wi fw similarli note wi includ world assert gener assert serial respect normal wb regular respect world ensur properti hold serial respect normal world ensur intent inconsist presenc properti satisfi conjunct properti argu desir properti algebra approach admit provid condit wi wi condit modest form regular behaviour valuat full condit satisfi hold achiev desir behaviour wi regular oe regular oe obtain properti condit satisfi wi wi world framework adopt properti affect properti intent decoupl semant belief intent classic set properti belief intent system properti formula oe valid inspect requir world evalu modal ensur satisfact wi wb note situat condit hold wb wi reduc condit classic counterpart properti oe ensur satisfact wi wb note wi wb condit satisfi classic counterpart assum serial belief show condit impli condit assum addit regular formula world properti shown desir properti interact discuss classic set satisfi world framework special reduc classic counterpart shown undesir properti intent present classic set avoid world framework affect properti belief intent plan logic framework model link intent action ensur logic properti int decomposit intent introduc end section extend languag enabl refer mean intend state achiev view plan recip program agent commit agent commit plan intent set regard plan complex combin primit primit action transit world combin plan primit action form oper achiev oper convert proposit oe plan express dynam logic style formula state oe true complet execut plan formula state oe true complet plan held point execut detail extend languag import note languag extens notion intent captur intent end intent plan achiev end formula state agent intend execut plan formula oe true agent world constraint intent execut plan agent intend intend agent achiev state intend properti captur oe agent intent independ plan intent form perform oe oe agent perform reason substitut intermediari end plan achiev intermediari end agent intend achiev oe oe intend plan achiev replac end mean intend execut plan oe relat work discuss number author recent defin logic intent konolig pollack explicitli address natur logic intent adopt call model intent emploi notion cognit structur int semant agent cognit state compon cognit structur relat intent set int proposit set world oe proposit denot world oe true iff int requir int close superset int requir candid int address problem distinguish success fail plan execut strictli evalu function structur close set intersect int int int obtain oe axiom base simpli oper framework import limit proposit equat set world true escap oe unclear extend framework ensur possibl desir pattern reason logic intent disjunct closur properti section support logic formula oe oe support logic formula oe oe isnt intuit arent complet clear distinctli undesir accept reject clear framework extend world framework explor shown extrem flexibl avoid arguabl undesir consequ inher normal modal show model theoret semant easili constrain desir properti support simpl restrict world subset world requir serial respect normal ensur belief intent interact arguabl manner restrict leav open option logic belief logic reflect intuit issu intend investig framework order obtain complet theori bdi specif incorpor tempor aspect attitud extend framework deal plan intent investig dynam intent reason action fail intend effect extens make semant framework refer bratman practic reason harvard cohen levesqu intent choic commit artifici konolig pollack theori intent joint confer artifici intellig page imposs world semant logic omnisci theori volum page acta rao georgeff asymmetri thesi problem linear time branch time intent logic twelfth intern joint confer artifici intellig recent unpublish work investig intent relat intent plan rao georgeff bdi theori practic intern confer system san inquiri mit gener world framework reason knowledg belief studia 
IR/1785 5 invert file dynam signatur file web directori fidel angel depart inform commun technolog de campu de spain web directori taxonomi classif web document kind ir system present specif type search document collect restrict area categori graph paper introduc specif data architectur web directori improv perform restrict search architectur base hybrid data structur invert file multipl signatur file variant base model hybrid architectur total inform hybrid architectur partial inform valid architectur mean develop variant compar basic model perform restrict queri special hybrid model partial yield posit load search system keyword web data hybrid data signatur invert file introduct ir system web retriev filter inform huge data base constitut www basic wai locat inform search web directori web directori ontolog ad possess search process combin navig improv qualiti obtain result search restrict document link area ontolog root node user navig paper examin detail restrict search characterist web point view data structur perform obtain perform improv hybrid data model invert file dynam signatur file basic model web directori consist basic compon vocabulari stand kei word index document directori categori structur repres hierarchi categori exist typic direct acycl graph small document file requir basic inform titl repres relationship invert file structur constitut present index techniqu perform huge data volum cat id kei basic info categori link categori file doc id basic info document file document link vocabulari categori link document link figur data structur basic model invert file categori graph invert file invert file figur show schemat data structur basic model web directori architectur base similar restrict search data model explain permit solv normal search process navig process effici invert file structur restrict search process categori graph area requir elabor access inform standard search carri retriev kei step consist determin result document belong graph area altern defin filter process altern consist obtain list document graph combin list result order obtain final result altern consist obtain categori list restrict area easier process obtain document check result list document locat node categori list solut solv effici search obtain great number result restrict wide graph area hybrid architectur problem restrict search li obtain invert list result undergo filter process base attribut base complex hierarch structur model data structur base dynam signatur file filter exact filter examin rest document model document signatur repres categori belong directli indirectli signatur file incorpor invert creat hybrid scheme invert signatur file invert list invert file local document identif signatur file categori code figur document identifi hybrid data structur signatur file invert list due possibl dynam gener signatur file queri combin oper invert list carri signatur file automat obtain document identifi built signatur categori local document identifi figur signatur file techniqu appli direct acycl graph web directori consist associ signatur categori node document gener signatur mean categori signatur directli indirectli code repres hierarch inform detail cat id doc id kei basic info basic info categori link categori file document file document link vocabulari categori link link document link sign doc sign doc sign doc signatur file figur data structur hybrid architectur invert file invert file categori graph invert file invert file implement variant base architectur establish implement carri consist develop web directori prototyp base real integr approxim categori distribut depth web directori classifi develop environ ultra machin processor mb memori gb store space hybrid model total inform hybrid model total inform correspond direct applic code techniqu categori graph categori signatur figur show dot line data structur size matter size due signatur special format document identifi global main repercuss size requir li index kei word impact smaller rest case system support categori million document hybrid model partial inform variant hybrid architectur aim reduc size signatur document reduc store space number oper reduc mean appli techniqu code level rest node inherit signatur genet code upper level regard requir store significantli approxim reduct requir space note index kei word document requir store space variant support million document limit number low level categori perform evalu work load situat consid high satur similar perform normal search prove restrict search evalu load level similar result obtain low medium load hybrid model improv perform approxim high load behaviour hybrid model vari figur perform variant total inform worsen significantli partial inform model improv time number result figur time basic hybrid total info hybrid info time number result figur time basic hybrid total info hybrid info time number result basic figur time hybrid total info hybrid info time number result figur time basic hybrid total info hybrid info paper describ hybrid data architectur invert file signatur design improv perform search restrict area categori graph variant architectur hybrid model total inform hybrid model partial inform variant improv perform respect basic model load situat system hybrid model total due bigger store space suffer dramat declin perform high load similarli basic model implement carri prove flexibl regard number document system regard number categori directori refer inform chapter addison weslei file versu signatur file text transact decemb anatomi hypertextu web search intern world wide web inform chapter pp addison weslei code repres hierarch inform web intern workshop web inform data manag acm cikm perform analysi signatur file acm analysi code american ontolog categori describ eighth intern confer inform knowledg pp text categor applic text ieee transact knowledg data volum pp search hierarch structur directori seventh intern confer inform knowledg manag 
HCI/532 4 proc ieee workshop human analysi hilton june ieee learn track human motion function analysi black hasti fleet dept stanford ca royal institut technolog sweden xerox palo alto research palo ca abstract present method model track human motion sequenc video imag analysi divid statist learn bayesian track estim statist model typic activ larg set human motion data human bodi repres set articul cylind evolut joint angl period motion develop set tool automat segment train data sequenc ident comput princip compon cycl algorithm account miss inform enforc smooth transit cycl learn tempor model prior probabl distribut human motion track adopt bayesian perspect approxim posterior distribut bodi paramet particl filter result algorithm track human subject monocular video sequenc recov motion complex unknown environ introduct model track human motion video import problem vari video sport interact human bodi approxim collect articul limb motion thought collect describ joint angl evolv time kei difficulti model bodi angl suitabl tempor primit prior statist analysi case repetit human motion motion sequenc natur sequenc ident cycl exact natur decomposit unknown model estim motion data present set tool carri identif automat suggest iter procedur gener segment respect ratio data align refer domain procedur princip compon individu cycl refer domain statist model technic difficulti includ miss inform motion necess enforc smooth transit cycl deal develop iter method function princip compon analysi base truncat fourier transform learn tempor model prior probabl distribut human motion bayesian framework track gener model imag likelihood observ imag data model gener model result posterior distribut repres close form posterior repres discret set sampl propag time particl filter prior distribut base pca serv improv effici particl filter constrain sampl region subspac paramet space result algorithm track human subject monocular video sequenc recov motion pose complex unknown background bayesian track detail summar unlik previou work align train paper length radiu global tran global rot figur human bodi consist set articul cylind freedom local coordin system axi direct limb joint angular express rel rotat bodi part tail autom method learn tempor model cyclic data form limit import class human motion learn method develop extens cyclic gener human motion relat work model human motion attract great deal attent comput graphic comput vision commun work comput vision commun recognit activ model human motion task hidden markov model extens recognit human action hmm typic provid approxim motion data good synthesi recognit brand recent work suggest fundament weak hmm model captur properti biolog motion smooth linear dynam model human motion repres explicit tempor curv describ chang time joint angl focu model joint angl track human motion simplest model place constraint smooth chang joint angl detail model repres explicit curv variat joint angl time model deriv biometr studi human motion learn data statist data function analysi detail common approach model motion curv align data activ variat subject model princip compon analysi curv data princip compon captur variat train set approach repres imag optic flow joint angl relat approach vector quantiz pca primari detail model track articul model peopl high human tempor curv constrain motion lie lower manifold bayesian method popular track human motion tempor curv thought provid prior probabl distribut valid human motion pca tempor curv statist model variat present train set construct probabilist prior work model human motion curv cyclic motion walk run cyclic motion simpl import class human activ extens studi current focu cyclic motion provid statist treatment ultim goal model human motion automat learn probabilist model train data involv automat segment data individu align activ model statist variat deriv probabilist model complic matter fact train data imperfect commerci motion captur miss data account model cyclic learn method enforc smooth transit cycl previou issu dealt manual case cyclic provid complet treatment automat cope miss data enforc smooth learn part develop model procedur period motion sequenc period motion repetit constitut natur unit statist model identifi train data prior build model segment carri manual procedur present align algorithm segment data automat base estim align cycl motion sequenc transform common refer princip compon transform cycl data comput statist model cycl interpret type specif motion princip compon main sourc deviat sequenc data set cycl statist construct prior distribut bayesian track motion data train form joint provid commerci system sequenc angl term type motion render subject observ altogeth data set consist motion sequenc render individu length motion sequenc rang frame denot length motion sequenc time index number angl motion sequenc angl motion sequenc written formal fz mg motion sequenc train data set sequenc indic set ft miss observ occur frequent data set marker occlud part motion captur system report angl posit coordin case typic frame prevent imput interpol valu neighbor observ spend effort design algorithm manner insensit artifact sequenc align describ procedur estim align paramet segment motion sequenc cycl motion estim base individu motion sequenc transform common analysi estim cycl simpli larg number candid valu assess qualiti align result simpl score function project index defin denot smallest integ greater equal origin sequenc domain pg denot valu project observ data refer coordin limb figur map written ji magnitud ji signal measur project sequenc combin singl defin ratio signal interpret variat data explain signal measur signal intens natur prefer valu produc larg ratio algorithm candid valu set maximum respect estim figur ratio repres set angl function candid period length show left shoulder elbow hip knee angl top seri correspond white signal bottom seri show accumul ratio minimum variat obtain ratio note magnitud normal produc estim figur show individu ratio subset angl accumul ratio function note sharp peak valu optim cycl length note ratio artifici gener white seri row approxim warrant approach respect chang valu fold procedur comput estim optim cycl length sequenc store valu arrai length arrai align multipl sequenc common domain rescal construct offset estim shift motion sequenc minim deviat common prototyp model analog previou paragraph exhaust search optim offset combin due high comput complex suggest iter procedur illustr figur comput approxim initi offset valu step initi offset valu function class minimum fit respect align data arg min updat offset paramet arg min perform improv goto step figur iter algorithm comput optim offset paramet defin refer signal step minim deviat respect align data offset sequenc minim predict error respect refer signal contrast exhaust oper requir comparison solut iter construct improv refer signal current offset signal turn improv offset estim repeat obtain iter optim algorithm termin improv fall threshold step predict clear algorithm converg figur align walk sequenc red curv denot repetit period regress spline estim converg figur show sequenc walk align procedur function class refer chose period constrain regress spline deriv spline design coincid boundari detail concaten refer signal figur discontinu transit cycl princip compon analysi break align data figur individu cycl comput princip compon result slice princip compon interpret major sourc variat data formul probabl model walk algorithm develop illustr figur algorithm address individu motion sequenc align figur sampl frequenc common domain due align paramet problem accommod step resampl comput function estim continu time step function regress spline case difficulti consist miss data design matrix simpli singular decomposit obtain princip cycl length mark miss valu end set time seri index set seri compet function estim domain arg min resampl data common refer imput miss observ stack obtain sequenc design matrix comput vector slice comput fourier coeffici store coeffici comput singular decomposit reconstruct design rank approxim appli fourier transform add obtain imput miss valu valu evalu jjx jj perform improv goto step figur function pca algorithm data imput ponent altern iter approxim scheme suggest recent hasti al context analyz gene express arrai altern svd step data imput step updat design matrix distanc addit comput svd directli princip compon obtain manner discontinu due data imput step explicitli enforc constraint accommod problem translat cycl fourier domain truncat coeffici comput svd fourier domain step reconstruct design matrix approxim step step reconstruct valu improv estim miss data repeat step improv estim iter process continu perform improv fall threshold converg algorithm proven argument similar section provid detail algorithm gener imput design singular decomposit output serv prior distribut track algorithm experi assum motion essenti superposit compon section bayesian track goal estim joint angl bodi pose sequenc imag time outlin bayesian framework defin gener model imag obtain likelihood term specifi probabl observ imag paramet bodi model singular decomposit formul bayesian prior distribut track reformul infer problem comput posterior distribut prior likelihood comput typic difficult high bodi model approxim motion term princip effect reduc arriv feasibl algorithm random vector summar rel joint angl time motion time interpret realiz model assumpt svd figur written form fourier column rearrang denot rearrang vector column coeffici map absolut time rel cycl posit denot speed motion mod bodi posit vector linear coeffici repres global translat rotat torso track person monocular video sequenc entail estim distribut time adopt bayesian perspect posterior probabl paramet time updat likelihood observ imag paramet posterior probabl previou instant tempor prior probabl distribut encod paramet chang time element bayesian approach summar detail reader refer gener imag model geometr optic model camera defin map scene coordin coordin system bodi model kinemat tree articul cylind bodi root global translat rotat map torso scene coordin rigid transform rel posit orient connect limb specif valu paramet vector valu defin set rel joint angl equat combin joint angl global translat defin configur bodi time camera model specifi model project imag geometr formul predict imag time function imag textur time model map surfac model model project textur back imag gener model imag time view map imag time model paramet time project model imag paramet time denot gaussian distribut standard depend view angl limb respect camera limb view obliqu tempor prior tempor model paramet describ bodi configur expect vari time express formal collect distribut individu compon jc jt empir determin standard deviat small scalar singular valu equat expect small vari walk cycl individu homogen transform matrix global bodi gaussian distribut impli gaussian distribut joint angl sampl distribut correspond bodi configur model distribut rel joint angl bodi likelihood model complet bayesian defin observ imag human model configur time base gener model predict imag time compar gener imag observ imag independ evalu gener model limb comput likelihood observ imag product result likelihood term limb pixel uniformli sampl project limb region defin likelihood observ imag imag angl limb princip axi imag plane camera function narrow view angl simpl model work formul robust likelihood model remain area current research limb complet larg likelihood low limb view narrow angl visibl surfac normal roughli perpendicular view true textur pattern highli distort limb thought occlud probabl view model occlud region introduc constant occlud limb occlud express likelihood mixtur imag likelihood occlud act visibl influenc actual imag angl limb princip axi imag plane likelihood imag likelihood limb defin imag occlud limb limb occlud likelihood observ imag bodi pose stochast optim posterior distribut due nonlinear likelihood function result viewpoint match ambigu posterior complic dynam model state evolut embodi tempor prior reason repres posterior weight set state propag time particl filter approach briefli describ method repres vector paramet posterior time repres sampl comput posterior time draw sampl posterior time shape paramet propag sampl point valu comput likelihood likelihood normal sum result set sampl approxim posterior distribut time experi illustr method show track walk person clutter scene ultra sparcstat implement ran rate approxim visual posterior distribut displai project model expect model ns ns likelihood sampl paramet initi gaussian prior time figur show track result frame sequenc show walk person note leg model align imag data arm due fact arm occlud prior driven leg part cycl larg occur model imag start drift person recov larger part bodi visibl paper describ fulli autom method learn period human motion train data statist method present detect length period segment optim align cycl present princip compon analysi techniqu build motion curv method cope miss data enforc smooth begin end motion cycl learn prior probabl distribut bayesian track framework track monocular imag sequenc perform particl filter demonstr result track person clutter imag sequenc acknowledg grate michael gleicher gener provid data experi discuss human motion refer cr dyer cyclic motion detect spatiotempor surfac curv pp bobick davi action brand shadow pp bregler learn recogn human dynam video sequenc pp bregler malik track peopl twist exponenti map figur track person sampl upper row show frame sequenc project expect model configur overlai lower row show expect configur frame lw campbel af bobick recognit human bodi motion phase space constraint pp rehg multipl hypothesi approach figur track pp blake track singular discontinu random sampl pp gordon approach bayesian state estim iee sonar program walk person pp isard blake contour track stochast propag condit densiti pp freeman bayesian estim human motion imag sequenc technic report mitsubishi electr research dynam bayesian network approach figur track learn dynam model pp silverman function data analysi springer human movement analysi base explicit motion model shah pp kluwer academ pub sm cr dyer affin invari detect period motion pp de la torr black framework model articul figur int conf autom face gestur pp fleet stochast track human figur imag motion ireland track person monocular imag sequenc wilson bobick parametr hidden markov model gestur recognit sept yacoob black model recognit activ tempor surfac hasti gener addit chapman tibshirani miss data gene express work depart stanford 
ML/2444 6 inform extract hmm shrinkag dayn freitag andrew mccallum research henri street pa abstract hidden markov model power probabilist tool model time seri appli success task part speech speech text segment topic detect paper describ applic hmm languag relat problem locat textual answer inform hmm state transit probabl word emiss probabl learn label train data machin learn lack sucient label train data hinder reliabl model kei contribut paper statist techniqu call signi cantli improv paramet estim hmm emiss probabl face train data experi seminar announc reuter acquisit shrinkag shown reduc error result hmm outperform system introduct internet make tremend amount text gener human vast quantiti inform easili manipul analyz comput inform extract process lling eld automat extract text exampl includ extract locat meet email extract acquir compani newswir articl compani paper advoc hidden markov model inform extract hmm type probabilist nite state probabilist tool model sequenc observ appli signi success includ tag speech recognit text segment topic detect al hmm foundat statist rich bodi establish techniqu learn paramet hmm train data classifi test data work hmm probabl word emiss probabl learn label train data case machin learn larg amount train data requir learn model gener high accuraci train data label dicult obtain small quantiti train data limit factor perform learn extractor kei contribut paper integr statist techniqu call shrinkag inform extract hmm shrinkag learn robust hmm emiss probabl face limit train data techniqu work paramet estim individu state estim calcul wai provabl optim condit shrinkag wide statist languag includ hmm acoust model speech recognit approach inform hmm form probabilist gener model entir document extract separ hmm built extract di erent eld document subset state distinguish word document determin gener state part extract recent report system break document small fragment learn predict approach requir unsatisfactori heurist choic result huge space fragment approach ectiv solv fragment problem mean viterbi ecient method nding probabl sequenc model state document previou project hmm inform di er shrinkag hmm state bikel al relat project di erent task learn hmm structur inform extract rosenfeld describ experi data seminar announc reuter newswir articl compani acquisit result show shrinkag consist improv perform absolut discount repres hmm outperform system extract task hmm inform extract task extract price document collect articl describ corpor acquisit imagin articl result stochast process involv unigram languag background model typic emit token model speci price typic emit token million gener process emit token background point switch price model return background model complet document slightli number special model process devot context model pre fragment xyz corp hmm repres kind process hmm nite state automaton stochast state transit symbol emiss automaton model probabilist gener process sequenc symbol produc start design start transit emit symbol select transit emit design nal state reach set fs probabl distribut symbol emiss vocabulari fw probabl state emit vocabulari item written state distribut set outgo transit probabl move state state written js model transit emiss probabl learn train data train sequenc sucient label uniqu path probabl calcul straightforwardli ratio count smooth ratio count model inform extract perform determin sequenc state gener entir extract symbol design state perform extract requir algorithm nding state sequenc hmm model sequenc symbol figur topolog naiv approach nding sequenc time exponenti sequenc dynam program solut call viterbi algorithm solv problem length sequenc number state viterbi algorithm execut time linear length ecient inform extract approach evalu number model inform extract hmm extract type eld multipl eld extract document separ hmm construct eld model entir requir segment document sentenc piec entir text train document train transit emiss probabl kind background state target state target state intend produc token extract fulli connect restrict transit structur captur context improv extract accuraci topolog set learn train data figur show topolog model figur simplest topolog expect context target state provid import clue search target text exploit clue ad sux figur target fragment vari token common begin end fragment object extract compani token end fragment attempt captur structur expand singl target state arrai parallel path vari length nal state longest path includ order account unusu long target fragment figur show set target path length figur illustr topolog paramet experi context window size target path count window size number pre sux state model figur window size requir model number pre sux assum simplic path count number sequenti target path path count model target vari length model figur path count model figur path count order train token document label part target text requir target state emit state emit token unambigu path topolog figur estim emiss probabl emiss vocabulari larg respect number train maximum likelihood estim emiss probabl lead poor word inappropri probabl prior conjunct maximum posteriori optim estim prevent estim improv estim optim paramet estim conjunct uniform dirichlet prior result wide laplac count word vocabulari increment singl extra occurr addit smooth altern smooth techniqu perform number word vari wide state state absolut discount method subtract xed word count greater result probabl mass distribut word count prior distribut case jz jv size number time occur train data state total number word occurr word train data state jz number uniqu word count state solut optim typic choic number uniqu word state count laplac smooth absolut discount calcul word distribut state train data state discuss method leverag word distribut relat state order improv paramet estim shrinkag machin learn task tension construct complex model state construct simpl model state complex model repres intric structur result poor paramet estim train data highli fragment simpl model result robust paramet perform poorli sucient express model data shrinkag statist techniqu balanc compet concern paramet estim state complex model estim relat state simpler model combin estim provabl optim condit shrinkag extens studi statist loui version techniqu wide statist languag model speech recognit emploi simpl form shrinkag combin estim weight learn weight expect maxim speech recognit form call delet interpol mercer shrinkag hmm inform extract shrinkag typic de ned term hierarchi repres expect similar paramet estim leav creat hierarchi de ne subset state word emiss distribut expect declar share common hierarchi word distribut figur show hierarchi pre state expect relat word ect fact simpler pre state repres singl state intern node hierarchi ect expect weaker similar group repres hmm emiss distribut simpl top hierarchi word uniform word vocabulari equal probabl uniform distribut includ longer context global uniform suffix prefix figur shrinkag con gurat address data sparsiti contextu show shrinkag state smooth local estim laplac absolut discount compar shrinkag hierarchi gurat distinguish class pre shrinkag con gurat absolut discount uniform absolut distribut shrunk uniform distribut global distribut target state shrunk common uniform state di erent parent hierarch figur target distribut handl global class pre shrunk parent pre sux parent shrunk share grandpar pre sux state shrunk singl share state target state state shrunk uniform distribut word probabl paramet estim leaf hierarchi linear interpol estim distribut leaf root local estim calcul train data maximum likelihood ratio addit train data intern node hierarchi union data children write word probabl estim node path start state estim uniform distribut root interpol weight estim written estim probabl word state written weight determin mixtur weight deriv empir optim ancestor state nding weight maxim likelihood hitherto unseen maximum found simpl form rubin word assum gener rst hierarchi node path probabl node word distribut gener word em maxim total likelihood choic node made word unknown em begin initi initi iter step calcul node predict word state js js deriv weight normal correct conceptu method make ineci train data carv set problem evalu individu word occurr held turn method similar commonli statist estim experi present result inform extract problem collect seminar announc post local newsgroup larg collect articl describ corpor acquisit reuter dataset problem de ned detail publish work perform algorithm measur document document task extract start time seminar assum singl correct answer present di erent time super cialli learner singl speaker locat window window window tabl ect perform chang size context seminar announc eld absolut discount speaker locat uniform global tabl ect perform di erent shrinkag con gurat seminar announc topolog window size singl target state predict identi es fragment repres start time learner predict align actual start identi ed human count error number test document learner predict correctli identi es instanc target eld divid number document learner issu predict recal divid number document instanc target eld present result term harmon assum context correct extract hmm exploit tabl explor ect perform window size extract task seminar announc domain absolut discount estim emiss probabl note smallest largest window size perform ect fact complex model train make dicult obtain reliabl paramet estim depend emiss estim rel shrinkag model make improv larger window size tabl show ect shrinkag perform topolog eld shrinkag gurat absolut discount case improv shrinkag reduc error speaker eld shrinkag eld attribut larg perform di erenc task rel infrequ eld time gener preval seminar announc half document distanc speaker locat tabl local mixtur weight pre path function distanc target state speaker locat uniform global tabl ect perform di erent shrinkag con gurat seminar announc topolog window size parallel erenti target path seminar end time bottom shrinkag con gurat due larg number spuriou extract document end time present make train state emit token havent shrinkag ect interest distribut mixtur weight vari function state role model tabl sampl run seminar announc weight local token distribut pre state shrinkag con gurat case note local weight declin distanc target intuit consist pattern closest local weight proport diculti ect result time eld tend occur predict context combin shrinkag multipl target path power tabl show ect shrinkag con gurat perform network tabl singl target state replac parallel target path length compar earlier perform locat eld perform hmm learn method combin shrinkag appropri design topolog yield learn algorithm base sound statist run perform level competit learn algorithm experi tabl compar hmm perform speaker locat srv hmm acquir statu srv hmm tabl srv repres hmm eld seminar announc corpor acquisit inform extract problem consist strong algorithm model question context window size target global shrinkag con gurat hmm obtain higher score srv perform uniform shrinkag con gurat beat srv tabl relat work hmm suit rang languag process task appli problem inform extract wai di er approach wai relat rosenfeld present ort learn hmm structur unlik approach singl hmm extract eld pack moder structur text research paper refer eld repres singl observ expect rel order eld structur learn challeng task leek appli hmm problem extract gene locat text contrast model leek model carefulli engin task gener topolog hierarch languag model individu state leek network topolog model natur languag syntax trim train exampl present model state unigram languag unclear smooth polici unknown token handl special state system al hmm perform extract de ned di erent eld extract model singl avoid result cult singl state target structur complet connect emiss transit probabl condit highli fragment probabl train data form shrinkag obtain robust paramet estim unlik averag di erent hmm averag distribut di erent emiss learn optim mixtur weight em paper demonstr abil shrinkag improv perform hmm inform extract tension desir complex model lack train data constant struggl machin learn shrinkag principl method strike balanc strive motiv high suitabl hmm inform extract task point avail viterbi algorithm avoid evalu number refer learn nder empir method data analysi chapman hall maximum likelihood incomplet data em algorithm journal royal statist societi machin learn inform extract inform domain phd carnegi interpol estim markov sourc paramet data pattern recognit robust tag hidden markov model comput speech languag automat speech develop sphinx system kluwer academ publish inform extract hidden markov model master uc san diego learn inform retriev phd univ massachusett cs tech report tutori hidden markov model select applic speech recognit ieee learn hidden markov model structur inform extract submit workshop machin learn inform extract van text segment topic track broadcast hidden markov model approach intern confer spoken languag process 
IR/125 5 topic local experi observ brian davison depart comput scienc state jersei nj usa juli abstract web page link relat content combin text possibl html anchor describ page foundat usabl web examin extent idea hold empir test topic local mirror spatial local page web find likelihood link page similar textual content similar sibl page link parent close anchor text repres part target anchor text discrimin unseen child page result present foundat success web includ search linkag intellig web agent introduct web page link relat content combin text possibl html anchor describ page foundat usabl web make user follow link link point relev content idea notic research implicit system servic found web todai idea basic case system fail mention explicitli cdr influenc measur paper attempt rectifi situat measur extent idea hold paper primarili address examin presenc textual overlap page relat issu qualiti descript web page relev web crawler search engin link primarili web human browser web shorter version paper confer paper user expect find page link text find page accur descript text present search engin result show empir evid topic local descript text repres target page find likelihood link page similar textual content similar sibl page link parent close anchor text repres part target anchor text discrimin unseen child page experi select set page web follow random subset link present page corpu measur textual similar nearbi remot page explor qualiti anchor link respect document describ motiv work give exampl includ web search rank crawler web prefetch describ present result conclud summari find suggest work motiv web structur small part order page link random glanc inspect gener show typic web page author place random link page except banner creat link page relat topic practic wide believ underli number system servic mention question describ web page common applic content web page situat titl descript page result page queri typic search text link page number system assum accur repres page includ system web index web index page web gener invert index page search popular search engin includ altavista lyco index sort perform function search engin index full text page www worm index titl anchor text index line text recent googl start index titl todai common major engin index titl page smaller servic research project intranet search engin opt reduc storag index common index html meta tag keyword descript search engin index text cite problem search engin spam author place keyword text relev current page design draw popular search index typic includ anchor text hypertext term repres page term describ page referenc except index anchor text googl present target page user redirect page text fact relat target page recent public queri return microsoft highest rank answer googl search engin address question anchor titl meta tag descript text repres target page text titl descript weight term text page goal provid evid make includ text addit target text index process search rank system search engin text analysi find page relev queri search engin incorpor addit factor user popular actual user link popular page link form page statu calcul link popular statu calcul assumpt page author link random page link author direct reader page interest relev topic current page link analysi approach clever depend set interconnect page relev topic interest richli interconnect order calcul page statu anchor text rank relev queri commun discov analysi laser demonstr linkag inform rank page comput textual propag relev backward link point relev page goal enabl engin find page good start point autom page dont rank highli base text analysi explain util anchor text show neighbor page topic engin engin metacrawl savvysearch dogpil search servic search index collect compil result search engin servic present result obtain attempt rank result perform addit process grouper perform result cluster inquiru fetch document analysi simpler version decid fetch page analysi inform provid origin search engin qualiti page descriptor import textual rank cluster page crawler crawler web crawler follow link expect relev client interest queri similar crawler result search engin start crawl web dataset assum find highli relev page local search start relev page dean similar approach find relat page crawler content current anchor text determin expand link examin nearbi page relev anchor text relev intellig agent varieti agent peopl web depend content page text anchor determin suggest user prefetch link user compar text neighbor estim relev page neighbor current find anchor text describ target page method data set initi data set page random set page select web web estim hundr million page entiti complet major search hundr million page fraction page retain dataset sampl web select random subset web open question progress made data set start point paper select random subset web randomli select page approxim million page local research search engin crawl index earli decemb page dataset time gener primarili result inquiri made major search engin hotbot page neighborhood result direct ancestor descend page select page dataset bia sampl page neighborhood page page page score highli queri search remain data set initi data randomli select outgo link page retriev page randomli outgo link page fetch page set test anchor text relev sibl page measur similar sibl page textual extract page retriev perl perl librari text extract html page perform custom code term replac punctuat whitespac term made strictli content text page includ titl meta tag includ alt text imag url extract perl librari custom code standard url format drop maxim match url titl descript bodi text anchor text target url anchor text includ text link surround text term span basic textual item term frequenc textual similar calcul perform textual straightforward describ section measur appli pair term refer word sourc document titl anchor gener document note measur produc score rang ident document gener score document term common gener score tfidf cosin similar calcul select widespread long histori inform retriev note idf valu calcul document combin retriev entir web specif formula pp time number doc number doc term document tfidf normal sum valu term document sum document cosin measur pp term probabl measur design measur likelihood term present target document simpli sum fraction queri term present target number time number term document overlap measur chosen measur amount overlap normal length calcul measur sum term smaller repres fraction experi perform primari experi perform includ measur textual titl descript page page children page random page page ancestor respect distanc parent document refer url anchor text page point anchor text random page anchor text page point link parent measur length descript provid descript meta tag anchor page textual content examin link page perform experi stop word elimin porter term stem similar result clariti graph case postpon appendix end paper featur select term count domain figur twenti common domain combin sort frequenc top ten domain au number page number link number page number link figur view distribut number link web page result gener characterist characterist dataset initi url retriev addit uniqu url retriev randomli fetch distinct child link page initi set fraction page length word figur distribut content length web page percent url match length figur distribut url match length similar top repres host repres url figur show frequent toplevel domain close half url url net approxim url repres home page url path compon initi dataset approxim link distribut shown figur respect content sampl distribut sourc target page present distribut initi dataset shown figur half web page word pair page link domain match time pair percentag random pair domain match time measur number segment match url score mean host port strict domain match point addit path segment match directori match addit distribut segment match length connect page shown figur figur show similar descriptor descript meta tag descript show poorer perform titl term suggest author includ term present page longer text descript find descript higher overlap tfidf score term prob titl desc overlap figur similar score compar text page comparison score graph length descript suggest page page characterist figur present similar score current page link random sibl subset link page metric demonstr random page text link page text common link page sibl page similar link page domain figur plot sibl page similar score function distanc refer url parent page distanc count number link link separ link distanc find closer url share term strikingli found present metric corrobor observ link page similar topic cluster parent page tfidf score term prob random sibl dom link diff dom overlap figur textual similar link random sibl link page link page domain comparison score graph anchor page characterist anchor length shown figur slightli lower averag report amitai titl length term shown figur text anchor term term term figur show anchor text score higher page metric similar anchor text page sibl target page score order magnitud random conflict highest score perform anchor text link page domain sourc case term probabl tfidf score anchor text vari amount surround text constant improv text small term probabl show declin addit word appar addit text provid lower likelihood present target page addit term addit word side chanc target page compar tfidf score term prob overlap num sampl distanc figur plot similar score sibl page function distanc refer url parent page term high neg correl chanc anchor text unlik overlap score figur show improv addit word potenti result compat report chakrabarti al found includ fifti byte text anchor catch refer term larg dataset link yahoo home page interpret addit text chanc import catch unimport lower term probabl score cancel result text surround anchor link text made term fraction anchor anchor length figur distribut number term anchor fraction page titl length figur distribut number term titl text web text web amitai examin linguist choic web author comparison document detail find similar characterist web page bigram popular raw typic bigram text reflect commerci web tabl list frequent bigram paper empir evid topic local page mirror spatial local web www page typic link page similar textual content found page significantli relat topic page page select nearbi page found evid topic local sibl page similar link parent closer found anchor text similar page sibl similar random score statist larg order magnitud suggest anchor tfidf score term prob random sibl dom link diff dom overlap figur perform anchor text link link text link text text sibl text random page comparison score graph mail click http www york home page web site contact web site web page updat world wide site inform unit state reserv page check copyright wide web tabl common bigram found remov bigram form verb text discrimin unseen child page note anchor text term found target page close titl term target titl overlap tfidf cosin similar score point averag text anchor improv similar measur shown anchor text high term probabl high tfidf impli page proxi repres part target page pitkow pirolli observ emploi provid semant linkag manner citat link tfidf score anchor text anchor text anchor text term prob overlap addit word figur perform vari amount anchor text link text document relat demonstr semant approxim textual measur present provid underpin web includ search linkag intellig web agent part futur plan extend analysi includ textual similar page type link model determin intern extern site link match host domain acknowledg due haym paul kantor discuss comment earlier draft member group work support part darpa order number onr contract refer robert dayn thorsten tom mitchel learn world wide web aaai spring symposium inform gather heterogen stanford march aaai press amitai hypertext import master edinburgh technic report amitai common hypertext link identifi phrasal descript target web document workshop hypertext inform retriev amitai anchor corpu analysi web page author convent lynn simon word web comput mediat commun intellect octob krishna bharat andrei broder techniqu measur rel size overlap public web search engin seventh intern world wide web april justin dayn thorsten joachim machin learn architectur optim web search engin aaai workshop inform august krishna bharat monika improv algorithm topic distil hyperlink environ intern acm sigir confer research develop inform page august sergei brin lawrenc page anatomi hypertextu web search engin seventh intern world wide web april marko balabanov yoav shoham collabor recommend commun march israel michael dan vladimir ur ad support dynam search eighth intern world wide web soumen byron david jon ravi prabhakar sridhar andrew tomkin mine web link structur ieee page august soumen byron piotr indyk enhanc hypertext categor hyperlink acm soumen byron prabhakar sridhar david jon kleinberg automat resourc compil analyz hyperlink structur text seventh intern world wide web april junghoo hector lawrenc page crawl url order seventh intern world wide web april soumen martin van den byron dom approach web resourc discoveri eighth intern world wide web brian davison adapt web prefetch workshop adapt system user model page posit paper publish comput scienc report dept mathemat comput eindhoven technolog brian davison topic local web annual intern confer research develop inform retriev juli brian wei wu appli link analysi web search poster eighth intern world wide web page daniel adel experi select search engin metasearch acm transact inform juli dean monika find relat page world wide web eighth intern world wide web page david jon prabhakar raghavan infer web commun link topolog acm confer hypertext hypermedia expand version adel daniel metasearch engin learn search engin queri ai monika allan michael marc najork url sampl ninth intern world wide web thorsten dayn tom mitchel tour guid world wide web fifteenth intern joint confer artifici page morgan august build mainten robot base internet search review current index data collect method prepar work packag eu telemat project septemb jon kleinberg authorit sourc hyperlink environ symposium discret algorithm page san januari expand version brian internet fish phd mit depart electr engin comput ai technic report mit artifici intellig laboratori brian internet fish construct kit sixth intern world wide web santa april steve lawrenc lee gile context page analysi improv web search ieee internet steve lawrenc lee gile neci meta search engin seventh intern world wide web april steve lawrenc lee gile access inform web henri lieberman agent assist web intern joint confer artifici intellig page august henri lieberman autonom interfac agent acm confer human factor comput march filippo menczer richard belew adapt retriev intern local context scale web machin longer version technic report san diego oliv mcbryan tool tame web intern world wide web filippo menczer adapt retriev agent heurist neighborhood inform discoveri intern confer machin learn mladen person implement design technic report depart intellig stefan univ octob martin porter algorithm strip karen sparck jone peter read inform retriev morgan san origin publish jame pitkow peter pirolli law electron frontier acm confer human factor comput march jason renni andrew mccallum web spider reinforc learn sixteenth intern confer machin learn erik selberg oren etzioni search comparison metacrawl fourth intern world wide web decemb erik selberg oren etzioni metacrawl architectur resourc aggreg web ieee tom spring search engin microsoft pc novemb onlin danni sullivan dr search engin novemb danni sullivan search engin featur search engin januari oren zamir oren etzioni web document feasibl demonstr confer research develop inform oren zamir oren etzioni dynam cluster interfac web search result eighth intern world wide web appendix section present result stop word elimin porter term stem figur match ident figur figur scale url match length stop word elimin url match length term stem figur distribut url match length similar titl desc stop word elimin titl desc term stem figur similar score compar text page random sibl dom link diff dom stop word elimin random sibl dom link diff dom term stem figur textual similar link random sibl link page link page domain distanc stop word elimin distanc term stem figur plot similar score sibl page function distanc refer url parent page term random sibl dom link diff dom stop word elimin random sibl dom link diff dom term stem figur perform anchor text link link text link text text sibl text random page 
HCI/683 4 peter softwar infrastructur ubiquit comput support synchron collabor heterogen devic ubicomp ubiquit comput springer lnc pp softwar infrastructur ubiquit comput support synchron collabor heterogen devic peter gmd german nation research center inform ipsi integr public inform system ambient workspac germani abstract ubiquit comput multipl user work wide rang devic user interact collabor multipl heterogen devic time configur devic chang frequent due highli flexibl mobil natur work practic produc requir architectur softwar infrastructur architectur design meet requir test architectur basi implement softwar infrastructur ubiquit comput environ function synchron cooper interact room element integr inform technolog experi current implement present introduct ubiquit comput environ offer wide rang devic size shape situat tight collabor user work synchron inform share devic due heterogen natur ubiquit comput softwar infrastructur provid user interfac advantag properti enabl tight collabor user work diff rent devic share devic develop softwar architectur offer flexibl extens devic part ubiquit comput environ architectur creat softwar system call basic environ activ collabor hypermedia beach softwar infrastructur environ support synchron collabor devic offer user interfac fit devic peter marriag summer peter requir form interact synchron collabor beach build share document access multipl interact devic concurr requir softwar infrastructur ubiquit comput environ support synchron collabor point relat work base architectur present set implement architectur show paper close overview experi idea futur work requir softwar infrastructur softwar infrastructur ubiquit comput environ multipl heterogen devic addit requir compar collabor softwar run distribut standard pc section explain requir organ interact devic form interact collabor user support wide rang devic integr devic environ support task hardwar configur give concret descript compon room element integr inform technolog compon mention develop context project similar environ meet context applic scenario interact devic ubiquit comput varieti interact devic compar desktop pc equip devic form support style aim provid requir form interact import softwar infrastructur open style interact extens futur develop current oper system platform offer direct support interact techniqu softwar infrastructur ubiquit comput environ integr devic driver interact style requir introduct interact model pen input dispatch singl point displai affect wide area speech gestur requir process level higher level influenc recognit step made lower awar ambigu type abstract defin easili map invoc function pen event type concept adequ interact techniqu hardwar button found offer similar function compar softwar button widget interact model kind event trigger directli devic driver flexibl map softwar function requir adapt visual due interact devic ubiquit comput displai broad rang size orient differ size devic scale select object user overview part document edit devic harder similar interfac devic interfac metaphor concept problem interact tabl orient output necessarili common user work user surfac posit tradit user simpli rotat sheet paper show interact user view object orient object prefer orient requir output analog input techniqu output devic visual displai found ubiquit comput environ rang output ambient displai visual common interfac provid low level describ visual approach idea separ model abstract physic ui depend output devic gener physic user interfac automat gener element interact model common interfac underli function collabor user support devic characterist properti ubiquit comput environ presenc collabor commun user devic requir interact ubiquit comput user access comput user leav interact chair walk larg public present surfac give present softwar detect quickli chang assign current user devic give user access privat instanc prepar materi present continu scenario bring case user give present access devic parallel public displai view privat annot addit electron devic simultan inform displai devic context influenc result view level privat relat adapt visual context defin devic contrast usag devic exampl devic digit concurr digit pc literatur pda access addit inform function wast space main displai public pda privat inform function relev user devic show inform offer function requir collabor devic situat complic case multipl user work standard method share edit wysiwi requir collabor user coupl workspac size devic cover displai size small larg fig interact remot document displai privat workspac access softwar tightli coupl compon view ensur user fit current work mode user work share workspac user overview workspac content zoom view work locat directli front user overview overview displai shrunk navig requir devic devic interact tabl wall offer challeng peopl devic interact simultan singl devic call singl displai groupwar addit multipl user displai collabor multipl user displai lead relat collabor user devic softwar run devic receiv event recogn input track concurr event sequenc exampl event sequenc draw stroke drag window technic user interfac design interact multipl user integr devic environ import area requir integr devic softwar work environ work context requir context awar softwar awar context act depend state surround environ common exampl context current locat devic specif user kind devic softwar applic run physic contextu inform current task project influenc behavior softwar softwar infrastructur maintain current context updat interfac sensor collect context inform context mechan exist inform applic similar input data collect sensor preprocess order gener inform layer abstract applic requir physic interact configur physic object meet room strongli depend current work mode made object trigger action softwar reflect adapt made user set due current collabor mode case state chang softwar essenti maintain consist part world small interact tabl assembl quickli yield larger homogen interact area desir support flexibl split subgroup softwar capabl dynam size format visual interact reflect possibl conceptu design user interfac support task interview studi carri found creativ team recur task softwar offer dedic select set extens meet futur requir gener function function common wide rang applic scenario function reusabl requir tailor function import exampl typic group task support creativ meet project task manag provid tailor modul concept capabl extend gener function chang exist code modul requir captur access inform recur task gain access gener inform ubiquit comput lot inform captur automat addit gener manual captur access dedic support hardwar configur restrict hardwar place requir softwar requir devic user perceiv devic consist individu hardwar compon compon fig run separ pc due limit hardwar smart board recogn singl pen posit configur segment receiv pen input user support user simultan give user impress homogen interact segment coupl softwar enabl multipl user collabor visual interact area spite limit hardwar physic space requir dynam configur meet independ problem identifi solv parallel team divid set solv problem defin amount team form solut present scenario show kind collabor mode support ubiquit comput demand configur devic dynam meet reflect design flexibl give team freedom work effici architectur softwar infrastructur respect requir previou architectur develop offer flexibl support heterogen devic ensur extens futur devic section overview layer model discuss layer duti detail architectur overview order provid reusabl compon hand hook adapt devic architectur organ horizont layer defin level abstract orthogon architectur structur model separ basic concern layer modul layer gener layer model layer core layer interact model physic model model tool model document model space fig softwar architectur horizont organ layer defin level abstract vertic model separ basic concern crucial synchron collabor space provid core layer enabl distribut access object level abstract architectur organ layer defin level abstract modul layer specif modul provid tailor function distinct task extend function defin gener layer gener compon provid basic function teamwork meet situat top layer consist compon perceiv lower layer introduc structur applic ensur extens model layer specifi basic structur top layer defin interfac user physic interact style common abstract devic implement basic concern separ architectur core layer offer special infrastructur make implement higher layer easier share object crucial distribut access multipl comput basic model ensur clear separ model user physic environ distinguish interfac abstract class implement basic model defin model part core layer structur concept document model defin base class function object part document tool model describ element directli attach user provid addit function document tool control work mode coupl user interfac model defin altern user interfac concept suitabl devic devic requir user interfac element part space enabl user interfac element distribut comput physic model relev part world top layer defin physic model interact devic object monitor adapt dynam sensor recogn support style interact interact model specifi interact style defin term describ part softwar describ user interfac conceptu level defin user physic model implement share object give user devic possibl access object interact model object local machin client adapt interact style local physic environ interact capabl exampl section connect interact model object architectur constraint mechan extens discuss model model basic separ concern aim model layer provid implement basic model basi implement higher layer import ensur extens gener compon modul interact model defin layer concept separ handl input output model layer base class view code updat view dispatch event specif control beach adapt version concept section document model separ domain model call domain applic model domain model repres entiti domain world applic model describ applic aspect present manipul domain object domain model includ string describ text text attribut font size applic model add edit state cursor posit select share edit state abil provid displai cursor user specifi coupl part edit state share user privat valu workspac applic model instanc rotat workspac user work interact tabl req properti tightli coupl gener gener collabor support import goal softwar system provid gener compon situat task illustr gener compon exampl beach elabor section basi document creat beach hypermedia data model gener document element includ input link basic object constitut inform workspac structur inform tool realiz gener layer beach toolbar document browser document browser special role defin connect user interfac part document shown offer possibl navig document workspac main element user interfac beach segment overlai complet visual interact area compon partit defin space document browser resid front segment background posit freeli similar window popular oper system toolbar smaller tool hand user time import part physic environ configur compon refer comput run beach client combin station composit compon current set share shown figur compon consist station station displai displai station belong compon combin displai repres complet interact area complet area fig displai ad remov displai view show immedi adjust size area due depend physic model view station displai fig displai area combin displai station homogen interact area provid adequ interact tradit interact extend support gestur written pen support audio output implement gener gestur event handl pen stroke drawn gestur recogn check similar set support gestur shape gestur event gener dispatch contrast refer specif gestur event stroke cross bound multipl view object dispatch gestur event implement capabl select control modul tailor support task gener element provid beach situat task specif support architectur modul modul add model element extend function exist compon provid hook core layer add toolbar modul plug beach chang exist code beach modul support creativ team collect idea brainstorm session idea gener session collect pda transfer public compon meet core special infrastructur aim core layer provid function make develop higher level conveni includ synchron access distribut automat updat depend event view devic sensor modul servic interfac focu main interest scope paper space order provid comput support synchron collabor platform distribut access share object section discuss properti groupwar framework highlight import featur softwar infrastructur ubiquit comput environ beach replic compon connect wireless network low bandwidth mbp share connect client initi increment share object reduc commun server synchron replic share object ensur persist minim coordin object group atom element replic transact consist spite concurr object ubiquit comput environ highli import ensur fast user interfac avoid delai wait optimist transact offer speedup conflict action harmless automat updat depend detect share object initi arbitrari comput varieti import mechan provid trigger updat automat state share object declar descript view kind output depend view attribut share object calli detect trigger attribut chang attribut workspac set view workspac open view chang devic similar constraint system amulet work distribut set beach client beach client beach cooper support beach client beach client wireless network network interact beach client beach client beach server persist store fig beach client run compon server event handl devic provid interfac hardwar capabl handl multipl user time devic multipl devic driver send tag identif beach altern standard event support adapt user interfac compon equip event gener devic driver assembl higher level event intuit draw stroke pen click pen event combin stroke gestur event gener depend shape drawn pen circl kind event strategi event dispatch strategi event receiv control keyboard event dispatch view pen posit event directli discard effect event track concurr event concept extend tracker object receiv event view hierarchi dispatch mechan interactor beach capabl handl tracker time map input devic id event devic view transform view orient size depend current core model replac standard handl adapt version support transform transform object respond messag transform point graphic primit imag transform appli wrapper object insert view hierarchi view transform chang similar idea introduc portal intern camera jazz devic configur illustr architectur devic configur section exampl exampl experi implement develop base architectur beach softwar infrastructur comp built characterist determin focu beach compon support pen finger main medium input beach direct visual interact compon perman part fig network connect aim support synchron cpu impli beach small devic pda devic perman connect network exampl show element gener layer layer defin concret class implement gener support compon devic collabor larg public smaller privat devic shown case conjunct tabl demonstr devic multipl user view prefer exampl focu share leav interact model configur show relationship share model interact model connect present combin multipl comput interact devic mention consist comput req attach smart board physic model defin compon fig consist station displai combin larg displai area smart board lab mount wall softwar chang set dynam board mobil equip sensor connect displai area divid segment displai larg workspac document browser mention applic model defin edit state function workspac physic model ui model displai show part displai area tool model document model fig consist combin form homogen displai area complet area show larg workspac tight collabor devic connect previou interest tight collabor implement show displai area separ segment public privat workspac displai consist displai case fig split segment segment show document browser privat workspac connect document browser shown enabl tight share browser result coupl navig applic model edit state share provid awar inform size segment interact model shown figur provid mechan displai public workspac techniqu scroll zoom physic model ui model tool model document model connect public workspac privat workspac fig split displai area segment connect document browser shown show privat workspac multipl user collabor devic combin enabl collabor user interact aim support user devic horizont displai requir orient user orient displai realiz approach user open overlai freeli move displai similar window fig overlai document browser workspac applic connect workspac interact model open view show workspac user rotat workspac prefer rotat workspac applic model physic model ui model tool model document model interact rotat rotat fig user work interact workspac separ browser separ applic user workspac prefer orient current state experi implement beach base architectur show provid function softwar infrastructur exist compon beach implement smalltalk space coast framework chosen version framework dolphin system experi current version beach prototyp intern comput fair march give interact present explor visitor instal environ special version beach german occup safeti health exhibit dortmund open public visitor experi work work ambient beach intern extern present discuss design meet coast open sourc requir identifi interest influenc design softwar architectur support form interact event dispatch adequ strategi offer flexibl free develop task defin action trigger control event extend event dispatch mechan good basi support multipl person devic concurr implement cooper applic top coast framework success conflict detect handl concurr action updat multipl distribut view carri automat framework case network queu transact transmit connect restor multipl user continu chanc rollback due inconsist time adapt visual rang devic similar characterist factor call user interfac import note adapt chang dynam enabl depend mechan gener visual share object space provid coast consequ separ applic domain model kei interact composit compon model implement share object store client easili restart crash resum stop combin share object space depend mechan model dynam collabor devic integr contextu inform provid sensor flexibl sensor attach local connect arbitrari comput sensor part state monitor state trigger action possibl includ modul ad function beach offer special task futur work develop beach navig access creat integr audio output issu integr mobil devic connect server devic restrict cpu memori pda approach realiz connect pda provid interfac system asynchron work long group work nonoverlap part feasibl practic unconnect essenti deal inconsist solut automat creation version object chang connect latest inform beach found acknowledg author colleagu student ambient discuss support time consum implement work norbert richard thorsten improv paper refer al teach learn multimedia classroom project acm confer multimedia pp chart futur research ubiquit comput acm transact vol march pp gd softwar engin issu ubiquit comput intern confer softwar engin acm pp instrument interact model design user interfac acm confer human factor comput system acm pp al zoomabl graphic sketchpad explor altern interfac physic journal visual languag pp extens zoomabl user interfac graphic toolkit java user interfac softwar technolog acm press user interfac architectur share editor singl screen acm siggraph symposium user interfac softwar pp homepag integr inform applianc interact ieee pp gener icon intern confer interact pp pda share public make person inform public inform person person pp march elsevi architectur implement java packag multipl input devic tech report comput scienc colleg md tangibl seamless interfac peopl bit atom acm confer human factor comput system acm pp user interfac paradigm smalltalk journal object orient program pp manag interfac design humanoid acm confer human factor comput system pp für die mobil für comput bad pp abowd provid integr support ambigu interfac acm confer human factor comput system acm pp van tailor domain object meet tool electron whiteboard acm confer comput support cooper work acm pp commun exampl mobil compon chi summari suit real integr architectur inform audio enhanc collabor electron white board intern confer auditori displai finland implement architectur support groupwar technic report human comput interact school comput carnegi mellon pa al collabor multipl pda connect pc acm confer comput support cooper work acm pp al amulet model effect user interfac softwar develop ieee transact softwar pp hudson pausch futur user interfac softwar tool acm transact vol march pp al design audio aura acm confer human factor comput system acm pp user interfac support creativ teamwork environ diploma darmstadt technic depart comput scienc multipl devic approach support interact acm confer human factor comput system acm pp user cooper environ consist multipl digit devic cooper build integr organ architectur tional workshop cooper build februari lectur note comput scienc pp context aid develop applic acm confer human factor comput system acm pp design objectori synchron groupwar coast acm confer comput support cooper work acm pp model collabor share object intern acm confer support group novemb pp coast ein für synchron groupwar present germani smart technolog homepag wysiwi earli experi interfac acm transact inform system pp singl displai model present collabor human factor comput system acm pp al interact landscap creativ innov acm confer human factor comput system pp integr meet support local desktop environ cscw acm pp cooper integr design architectur space inform space al cooper build integr organ architectur intern workshop cooper build februari lectur note comput scienc pp für und der pp gener interact base integr design real virtual world carrol interact addison pp facilit explor interfac design altern humanoid model interfac design acm confer human factor comput system pp model groupwar support synchron collabor heterogen devic sep ieee cs press al dynam coupl displai flexibl creation share workspac user interfac softwar technolog nov acm press plastic user framework research agenda interact pp comput scienc issu ubiquit comput commun pp