Instructor: Prof. Elena Zheleva
Time: TTh 3:30-4:45pm
Location: Virtual Classroom
Office hours: Wed 2:30-4pm
Two roads diverged in a yellow wood,
And sorry I could not travel both [...]
I took the one less traveled by,
And that has made all the difference.
"The dramatic success in machine learning has led to an explosion of
AI applications and increasing expectations for autonomous systems that exhibit human-level intelligence. These expectations, however,
have met with fundamental obstacles that cut across many application areas. [... One of the obstacles] concerns the understanding of cause-effect connections. This hallmark of human cognition is [...] a necessary (though not sufficient) ingredient for achieving human-level intelligence.
This ingredient should allow computer systems to choreograph a parsimonious and modular representation of their environment, interrogate that representation, distort it by acts of imagination and finally answer "What if?" kind of questions. Examples are interventional questions: ΄What if I make it happen?‘ and retrospective or explanatory questions: ΄What if I had acted differently?‘ or ΄what if
my flight had not been late?‘ Such questions cannot be articulated, let alone answered by systems that operate in purely statistical
mode, as do most learning machines today."
--From "The Seven Tools of Causal Inference with Reflections on Machine Learning" by Judea Pearl
Causal reasoning is an integral part of data science and artificial intelligence. The goal of the course on Causal Inference and Learning is to introduce students to methodologies and algorithms for causal reasoning and connect various aspects of causal inference, including methods developed within computer science, statistics, and economics. The course will cover state-of-the-art research on causal reasoning and prepare students to conduct research in this area.
This is a seminar course. The goal of the course is to expose graduate students to state-of-the-art research on causal inference. The class project plays a central role in the course, and it should be taken as an opportunity to connect your research area of interest to the course topics.
Approximately one third of the course will be lecture-based using the following book by Judea Pearl, Madelyn Glymour and Nicholas Jewell: Causal Inference in Statistics: A Primer (Wiley Press 2016). The rest of the course will be student-led presentations of recent papers from the growing body of causal inference research.
The class will meet synchronously on Zoom (link posted on Piazza). We will be using Piazza for all course discussions and materials. Students registered for the course will be sent an enrollment email before the first day of class.
Written paper summaries
Research paper presentations
Course project (proposal, progress report, final presentation, final report)
CS 412 Introduction to Machine Learning or consent of the instructor.
Check Piazza for an up-to-date schedule.
Judea Pearl. The Seven Tools of Causal Reasoning with Reflections on Machine Learning. Communications of the ACM. 2019.
Optional: Brenden M. Lake, Tomer D. Ullman, Joshua B. Tenenbaum, Samuel J. Gershman. Building Machines That Learn and Think Like People. 2016.
|9/1||Hypothesis testing and randomized controlled trials||CIT Ch. 11 PTDS Ch. 18||Professor|
|9/3||Probability and statistics: review||CISP Ch. 1.1-1.3||Professor|
|9/8||Structural causal models||CISP Ch. 1.4-1.5||Professor||Form a team by 9/9 and post it @5|
|9/10||Graphical models||CISP Ch. 2||Professor|
|9/15||Interventions, adjustment formula back-door criterion||CISP Ch. 3.1-3.3||Professor||HW 1 out: 1.5.3 2.4.1 c) 3.2.1 c) 3.3.3 a)b)c)|
|9/17||Front-door criterion, covariate-specific effects, inverse probability weighing||CISP Ch. 3.4-3.6||Professor|
|9/22||Mediation and causal inference in linear systems||CISP Ch. 3.7-3.8||Professor|
|9/24||Defining and computing counterfactuals||CISP Ch. 4.1-4.2||Professor||HW 1 due 9/25 11:59pm|
|9/29||Counterfactual probabilities, counterfactuals in linear systems||CISP Ch. 4.3||Professor||Project proposal due 11:59pm Specs: @29|
|10/1||Counterfactuals: attribution, mediation, practical uses||CISP Ch. 4.4-4.5||Professor||HW 2 out: 3.4.2 (re-frame and re-use the data from Table 3.2) 3.8.1 f)g) 4.3.2 b) 4.5.2 a)|
|10/6||Do calculus and transportability||
Main paper: Bareinboim, Pearl. Causal inference and the data fusion problem. PNAS 2016.
Bareinboim, Pearl. External validity: From do-calculus to transportability across populations. Stat. Science 2014.
|Professor||Presentation and participation rubrics|
Main paper: Bareinboim, Tian, Pearl. Recovering from selection bias in causal and statistical inference. AAAI 2014. (best paper award)
Zhang, Gong, Scholkopf. Multi-source domain adaptation: A causal view. AAAI 2015.
Main paper: Mohan, Pearl, Tian. Graphical models for inference with missing data. NIPS 2013.
Mohan, Thoemmes, Pearl. Estimation with Incomplete Data: The Linear Case. IJCAI 2018.
|Ellen, Siham||HW 2 due 10/13 11:59pm|
Main paper: Heinze-Deml, Maathuis, Meinshausen. Causal structure learning. ARSA 2018.
Spirtes, Zhang. Search for causal models. Chapter 18 of Handbook on graphical models. 2018.
|10/20||Networks: Interventions under interference||
Main paper: Fatemi, Zheleva. Minimizing Interference and Selection Bias in Network Experiment Design. ICWSM 2020.
Fatemi, Zheleva. Network experiment design for estimating direct treatment effects. MLG 2020.
|Guest speaker: Zahra Fatemi|
|10/22||Networks: SCM for interference||
Main paper: Ogburn, VanderWeele. Causal diagrams for interference. Statistical science. 2014.
|10/27||Networks: Homophily vs. contagion||
Main paper: Shalizi, Thomas. Homophily and contagion are generically confounded in observational social network studies. Sociological Methods and Research, 40, 2011.
Shalizi, McFowland III, Controlling for Latent Homophily in Social Networks through Inferring Latent Locations 2016.
|10/29||Networks: Relational models and causality||
Main paper: Sherman, Arbour, Shpitser. General identification of dynamic treatment regimes under interference. AISTATS 2020.
Arbour, Garant, Jensen. Inferring network effects in observational data. KDD 2016.
|Guest speaker: David Arbour, Adobe Research|
|11/3||No class (Election Day)|
|11/5||Networks: Chain Graphs||
Main paper: Sherman, Shpitser. Identification and Estimation Of Causal Effects from Dependent Data. NeurIPS 2018.
Shpitser. Segregated Graphs and Marginals of Chain Graph Models. NIPS 2015.
|Jason, Christian||Progress report due 11:59pm|
|11/10||Matching and propensity modeling||
Main paper: Liu, Dieng, Roy, Rudin, Volfovsky. Interpretable Almost Matching Exactly for Causal Inference. AISTATS 2019.
Shahid, Zheleva. Counterfactual learning in networks: An empirical study of model dependence. AAAI-WHY 2019.
|11/12||Heterogeneous treatment effects I||
Main paper: Pearl. Detecting latent heterogeneity. Sociological Methods & Research 2015.
Tran, Zheleva. Learning triggers for heterogeneous treatment effects. AAAI 2019.
|Guest speaker: Chris Tran|
|11/17||Heterogeneous treatment effects II||
Main paper: Alaa, Schaar. Limits of estimating heterogeneous treatment effects. ICML 2018.
Künzel, Sekhon, Bickel, Yu. Metalearners for estimating heterogeneous treatment effects using machine learning. PNAS 2019.
Main paper: Ovaisi, Ahsan, Zhang, Vasilaky, Zheleva. Correcting for selection bias in learning-to-rank systems. WWW 2020.
Singh, Joachims. Fairness of exposure in rankings. KDD 2018.
|Guest speaker: Zohreh Ovaisi|
Main paper: Magliacane, van Ommen, Claassen, Bongers, Versteeg, Mooij. Domain Adaptation by Using Causal Inference to Predict Invariant Conditional Distributions. NeurIPS 2018.
Mooij, Magliacane, Claassen. Joint Causal Inference from Multiple Contexts. JMLR 2020.
|Guest speaker: Sara Magliacane, University of Amsterdam, MIT-IBM Watson AI Lab|
|11/26||No class (Thanksgiving)|
|12/11||Finals week||Final project due 3pm|