Graphical Models
and
Causality
This class is a weekly reading group discussing research papers on causality inference from observational or experimental data. In a purely observational setting, quantities of interest (variables) can be recorded, but not acted upon. In an experimental setting, some controllable variables can be acted upon. The selected papers aim at understanding machine learning techniques to infer causality, including causal graphs derived from "graphical models”.
Tuesday 17:0018:00PM, CAB H 57 (plan)
Isabelle Guyon
Pattern Analysis and
Machine Learning Group
Office:
CAB E 45 Universitätstrasse 6
Phone: (044) 63 23010
E m a il: guyoni@inf.ethz.ch
Joachim Buhmann
Pattern Analysis and Machine Learning Group
Office: CAB G 69.2 Universitätstrasse 6
Phone: (044) 63 23124
E m a il: jbuhmann@inf.ethz.ch
Prerequisites
These are recommended but will
not be strictly enforced.
Each week, the student will have to read a paper or a book chapter, which will be discussed in class. The students are encouraged to familiarize themselves with Bayesian networks using the Genie software. The software works directly on Windows platforms, but may also be run through Wine on Unix.
This is tested on Ubuntu Linux 5.10:
* Install wine (http://www.winehq.com/). Most distributions should
have their own way to install this package, e.g. "aptget
install wine" on debian based distributions.
* Download setup file of genie and run: "wine genie2_setup.exe",
this should install all the needed files to
~/.wine/drive_c/Program Files/GeNIe 2.0
* change to this directory and run: "wine genie.exe"
The class is worth 4 credit points
Requirements:
Each student will have to take the lead of the discussion on one paper. Students should sign up for a date. This will be done on a firstcome firstserve basis. Please email the instructor guyoni@inf.ethz.ch. This assignment will require a little more preparation that particular week, including writing a summary of the paper and a log of the discussion. This is NOT a presentation: slides may be used as partial support, but are not required.
Grading:
There will be no exam. Attendance to at least 2/3 of the reading groups is required. There will be a signup sheet every time to check presence and completion of the homework assignment (usually reading a paper or book chapter). You earn one point for presence and one point for homework completion. You must earn at least 18 points to get your credit points for the class.
Schedule
The readings are organized around 4 themes revolving around book chapters from Judea Pearl's book "Causality". They are complemented by algorithm/theory papers and book chapters of other authors presenting a complementary viewpoints and application papers. This tentative list of papers is subjet to change, please check every week the assignement for the following week. Copies of the paper to read the next week will be provided. Send requests to guyoni@inf.ethz.ch.
Dates (Tuesdays) 
Theme 
Paper  Discussion leader  Summary, discussion, and/or slides  
1 
4 April 
Introduction 
Graphical models and causality
reading group presentation (tutorial) 
A. Elisseeff 
Slides [ppt]
Slides [pdf] 
2  11 April 
I. Basic concepts 
Bayesian
models without tears, Eugene Charniak 
I. Guyon 
Install Genie. Build
an example network with the software implementing the example
of Fig. 2 of the paper. Slides. Summary. 
3  18 April 
Probabilistic Reasoning. Chapter 14 of the book "Artificial Intelligence:
A modern approach" by Stuart Russell and Peter Norvig. 
Jiwen
Li 
Summary. Slides
a. Slides
b. 

4  25 April 
Introduction to probabilities,
graphs, and causal models. Chapter 1 of the book "Causality" by Judea Pearl.
The whole
chapter is available in pdf. 
Severin Hacker 
Summary. Slides 

5  2 May 
II. Basic methods 
Belief propagation. An
introduction to factor graphs, by HansAndrea Loeliger. 
HansAndrea
Loeliger 
Slides. 
6 
9 May 
Structure learning. A tutorial on
learning with Bayesian networks, by David Heckerman. 
Markus
Kalisch 
Summary. Slides
[ppt]. Slides [pdf]. 

7  16 May 
Variational methods. An
Introduction to Variational Methods for Graphical Models,
by Michael Jordan, Zoubin Ghahramani, Tommi Jaakkola, and
Lawrence Saul. 
Patrick Pletscher 
Summary. Slides . Slide handouts. 

8  23 May 
A theory of inferred causation.
Chapter 2 of the book "Causality" by Judea
Pearl. (see also availability from Pearl's site) 
Daniel
Küttel 
Slides (Pearl).
Slides (Daniel).
Summary 

9 
30 May 
III. Identification of causal
dependencies 
Statistical causality analysis
of INFOSEC alert data Xinzhou Qin and Wenke Lee 
Annie Chen 
Slides. 
10 
6 June 
Causal diagrams and the identification
of causal effects. Chapter 3 of the book "Causality" by Judea Pearl. (see also availability from Pearl's site) 
Saikumar
Chalasani 
Slides (Pearl).
Slides (Saikumar). 

11 
13 June 
Application. Using
Bayesian networks to analyze expression data Nir Friedman, Michal Linial,
Iftach Nachman, and Dana Pe'er 
Andreas
Kägi 
Slides. Summary. 

12 
20 June 
IV. Control, action, planning 
The art of science
of cause and effect. Epilogue of the book "Causality" by Judea Pearl. 
Ulf
Holm Nielsen 
Slides 
13 
27 June 
N1 experiments suffice to
determine the causal relations among N variables, Frederick Eberhardt, Clark
Glymour, Richard Scheines 
Simon Meier 
Slides. Handouts. 

14 
4 July 
Application. Marginal
structural models and causal inference in epidemiology,
by James Robins, Miguel Angel Hernan, and Babette
Brumback. 
Thomas
Fuchs 
Slides 
Links
Books:
1) Bayesian networks and Decision graphs  FB Jensen.
2) Bayesian artificial Intelligence  Kevin B. Korb and Anne K. Nicholson
3) An introduction to Graphical models  Kevin P. Murphy (http://www.cs.ubc.ca/~murphyk/Papers/intro_gm.pdf)
4) Bayesian networks and beyond  Unpublished book Daphne Koller and Nir Friedman
Software:
1) Hugin (http://www.hugin.com) is an excellent tool to construct and test Bayesian networks.
2) WinBUGS (http://www.mrcbsu.cam.ac.uk/bugs/winbugs/contents.shtml) is a statistical tool to estimate the Bayesian inference with MCMC simulations using Gibbs Sampling.
Ebooks from the ETH library:
*Applied Bayesian modelling. Peter Congdon  Wiley (2003) (http://www3.interscience.wiley.com/cgibin/booktoc/104531773?CRETRY=1&SRETRY=0 )
*Bayesian approach to image interpretation. Sunil K. Kopparapu, Uday B. Desai  Kluwer Academic Publishers (2001) (http://ebooks.springerlink.com/summary.asp?id=70076 )
*Bayesian approaches to clinical trials and healthcare evaluation. David J. Spiegelhalter, Keith R. Abrams, Jonathan P. Myles  Wiley (2004) (http://www3.interscience.wiley.com/cgibin/booktoc/107614005 )
*Bayesian artificial intelligence. Kevin B. Korb, Ann E. Nicholson  Chapman & Hall/CRC (2004) (http://www.statsnetbase.com/books/1219/c3871_fm.pdf )
*Bayesian economics through numerical methods: a guide to econometrics and decisionmaking with prior information. Jeffrey H. Dorfman  Springer (1997) http://ebooks.springerlink.com/summary.asp?id=99654 )
*Bayesian forecasting and dynamic models. Mike West, Jeff Harrison  Springer (cop. 1997) (http://ebooks.springerlink.com/summary.asp?id=104512 )
*Bayesian nonparametrics. J.K. Ghosh, R.V. Ramamoorthi  Springer (2003) (http://ebooks.springerlink.com/summary.asp?id=98937 )
*Introduction to Bayesian statistics. William M. Bolstad  Wiley (2004) (http://www3.interscience.wiley.com/cgibin/bookhome/109855377 )
*Likelihood, Bayesian, and MCMC methods in quantitative genetics. Daniel Sorensen, Daniel Gianola  Springer (2002) (http://ebooks.springerlink.com/summary.asp?id=98912 )
*Measurement error and misclassification in statistics and epidemiology: impacts and bayesian adjustments. Paul Gustafson  CRC Press Company (2004) (http://www.statsnetbase.com/books/1196/c3359_fm.pdf )
*Multivariate Bayesian statistics: models for source separation and signal unmixing. Daniel B. Rowe  Chapman & Hall/CRC (2003) (http://www.statsnetbase.com/books/980/c3189_fm.pdf )