Stat 521A Spring 2009 (term 2)
Topics in multivariate analysis: Probabilistic graphical models

• Join groups.google.com/group/stat521a-spring09 email list
• Office hours, CS 187, Fri 10-11 am
• Grading. There will be weekly homework (theory and matlab programming exercises), and a class project. The grade will be 40% homework, 60% project.
• Textbook. We will use a draft of "Probabilistic graphical models: Principles and Techniques", by Daphne Koller and Nir Friedman, to be published by MIT Press in Fall 2009. Students can buy a copy of (parts of) this book from Copiesmart. Click here for their table of contents (password required - email me).
  In addition, I will handout a few chapters from another draft book, "An introduction to probabilistic graphical models", by Michael Jordan.
• Workload. We will cover about 1 chapter per lecture. The lectures will summarize the salient points of each chapter. There is no exam, so you only need to read enough to be able to do the homeworks, and to do an interesting project. However, you will get the most out of the class if you keep up with all the reading. Please expect to spend about 15h/week on the class (outside of lectures).
• Pre-requisites. Prior knowledge of graduate level machine learning (eg CS540) and/or multi-variate statistics (eg Stat 541C) is required. Send me email if in doubt.
• All lecture slides (836 pages!)

Tentative Timetable

L#	Date	Topic	Reading	Homework
L1	Tue Jan 6	Intro, directed graphs	K&F Ch 3. Optional review / survey paper: Graphical models M. I. Jordan. Statistical Science, 19, 140-155, 2004.	.
L2	Thu Jan 8	Undirected graphs	K&F Ch 4. Optional: Markov properties of graphs, chapter from Jordan's book.	.
L3	Tue Jan 13	UGMs, CRFs	Ch 4	hw1.pdf. hw1Sol.pdf.
L4	Thu Jan 15	Structured CPDs, Template models	Optional: Ch 5, Ch 6	.
L5	Tue Jan 20	Gaussian networks	Ch 7. Optional: Multivariate Gaussians Jordan ch 13.	.
L6	Thu Jan 22	Expfam	Ch 8	.
L7	Tue Jan 27	VarElim, Clique trees	Ch 9, Ch 10	.
L8	Thu Jan 29	Message passing on (Clique) trees	Ch 10, my HMM message passing chapter, Rabiner, Proc. IEEE, 1989 A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition,	hw2.pdf, hw2Sol.pdf
L9	Tue Feb 3	Loopy belief propagation and friends	11.1-11.3	.
L10	Thu Feb 5	EP, mean field and variational Bayes	11.4-11.5	.
L11	Tue Feb 10	Sampling	Ch 12	hw3.pdf hw3Sol.pdf
L12	Thu Feb 12	No class (SD)	.	.
L13	Tue Feb 17	Spring break	.	.
L14	Thu Feb 19	Spring break	.	.
L15	Tue Feb 24	No class (Duke)	.	.
L16	Thu Feb 26	Guest lecture by Brent Boerlage, founder of Norsys, a Vancouver-based company that sells a popular Bayes net package called Netica	.	.
L17	Tue Mar 3	MAP estimation	Ch 13	hw4.pdf, familyTree.zip (matlab data)
L18	Thu Mar 5	Continuous & discrete nodes	Ch 14, FHT ch17	.
L19	Tue Mar 10	Inference in dynamical systems	Ch 15	.
L20	Thu Mar 12	Parameter estimation in DGMs	Ch 16, 17	.
L21	Tue Mar 17	Structure learning in DGMs: statistics	Ch 18	.
L22	Thu Mar 19	Structure learning in DGMs: algorithms	Ch 18	hw5.pdf, hw5Data.zip (matlab data)
L23	Tue Mar 24	Parameter learning in DGMs with missing data	19.1-19.3	.
L24	Thu Mar 26	Structure learning in DGMs with missing data	19.4-19.7	.
L25	Tue Mar 31	Param learning in UGMs	20.1-20.6; FHT ch17	hw6.pdf
L26	Tue Apr 7	Struct learning in UGMs	20.7-20.9	.

List of most important topics