EECS 584 --- Artificial Intelligence 2

Instructor: Barbara Di Eugenio

Time: MW 6:30-7:45 pm
Location: LH 301

(The full syllabus (in postscript) can be found here. Note that the syllabus is periodically updated)

Machine Learning     Natural Language Processing     Planning         Assignments    

We will adopt the most current approach to Artificial Intelligence, that sees it as the study of intelligent agents. We will explore the application of AI techniques to information technology and the Web. E.g., intelligent agents are used to search the web for useful information, as a foundation for computer supported collaborative work or for electronic commerce.

After briefly reviewing some foundations of AI such as search algorithms, we will spend the rest of the semester on some of the most exciting research areas these days, such as:

• Machine Learning and Data Mining ([Mitchell 97]). Applications range from programs that discover general rules in large data sets, to information filtering systems that automatically learn users' interests, to automatic classification of web pages.

  See below for references to the papers we will discuss in class; see the syllabus linked above for the detailed schedule.
  
  

• Natural Language Processing ([Jurafsky and Martin 00], [Allen 95], [HLTSurvey 96], [NLP-WWW 97]). Among the most promising applications of NLP technology today are: information retrieval from the web, automatic topic detection and summarization of documents and newspaper articles, machine translation, and spoken dialogue systems.
  
  
• Planning. Applied among others to the construction of Internet softbots (see eg the Intelligent WebWare project at the University of Washington).

We will explore each topic by starting with the relevant chapters in one of the required books ([Mitchell 97] and [Jurafski and Martin 00]), that provides the foundations for building intelligent agents. We will supplement the books with other readings. We will experiment with the techniques we will study and apply them to problems of interest to the students.

References

Tom Mitchell. Machine Learning, McGraw Hill, 1997.
Daniel Jurafsky and James Martin. Speech and Language Processing. Prentice Hall, 2000.

Other readings

• Stuart Russell & Peter Norvig. Artificial Intelligence, a Modern Approach. Prentice Hall, 1995.
• James Allen. Natural Language Understanding --- 2nd Edition. The Benjamin/Cummings Publishing Company. 1995.
• Natural Language Processing for the World Wide Web. AAAI Spring Symposium. 1997
• Survey of the State of the Art in Human Language Technology. 1995.

Prerequisites

Machine Learning pointers and papers

General resource pages for Machine Learning:

• Two comprehensive lists of resources (research groups, bibliographies by topic, software, etc):
  • at the University of Wien, Austria
  • David Aha's ML page
• A repository of data sets, at the University of California at Irvine

After covering the basics of ML from Mitchell, we will discuss the following applications of ML to WWW and Internet:

• Learning User's interests regarding:
  
  
  • web sites: WebWatcher from Carnegie Mellon University (Joachims et al. 97), and various systems
    developed by the Machine Learning group at the University of California at Irvine (Pazzani and Billsus 97; Ackerman et al. 97)
    
    
  • books and documents: The Recommender system, University of Texas at Austin (Mooney and Roy 00)
    
    
• Extracting Knowledge from the web (i.e., classifying web pages): WebKb from Carnegie Mellon University (Craven et al. 97).
  
  
• Text mining from text: The DiscoTEX system, University of Texas at Austin (Nahm and Mooney 00)
  
  

Natural Language Processing pointers and papers

Planning pointers and papers

Assignments

Homework 1 (due ...)

Installation instructions

The program for ID3 (courtesy of Tom Mitchell, Carnegie Mellon University) is written in Common Lisp. We are using the GNU implementation of Common Lisp, GCL, available on the EECS servers at /usr/local/bin/gcl. You can use another Common Lisp, e.g. on your PC, if you so wish.

You need the following three files:

• decision-tree.lsp: contains the code
• breast-cancer-data.lsp: 699 data points on breast cancer detection, courtesy of University of Wisconsin Hospitals, via the UCI data repository (the data has been reformatted to be consistent with the ID3 implementation we use)
• breast-cancer-names: describes the meaning of attributes (plus specifies attribution for the data)

Homework 2 (due ...)