Graphical Models

CS 395T/SDS 386C, Fall 2014
GDC 1.406, Mon & Wed 9:30 - 11:00

Home | Instructor | Syllabus | Homeworks | Readings

Overview Probabilistic Graphical Models provide compact and analytically useful representations of joint distributions over a large number of variables, using graphs. Each graph represents a family of distributions -- the nodes of the graph represent random variables, the edges encode independence assumptions, and weights over the edges and cliques specify a particular distribution within the family. There are two main classes of tasks within this framework: the first is to perform inference, given the graph structure and parameters; and the second is to learn the graph structure and parameters themselves from data. The course will cover these fundamentals of probabilistic graphical models, as well as related specialized topics, such as relational and template graphical models, dynamic graphical models, and causality.

Grading Three problem sets (75 % of final grade), and readings/presentations (25 % of final grade).

& Papers
Graphical models, exponential families, and variational inference. M. J. Wainwright and M. I. Jordan.
Foundations and Trends in Machine Learning, Vol. 1, Numbers 1--2, pp. 1--305, December 2008.

Graphical Models with R. Soren Hojsgaard, David Edwards, Steffen Lauritzen.

Probabilistic Graphical Models: Principles and Techniques. D. Koller and N. Friedman.

Homework Policy Each student is expected to submit an individually written homework. When using information from papers, or other external sources, please cite this information. The readings will be in teams of size 3 or 4.

Grading Policy Three Homeworks (75 %).
Readings (25%).

The homeworks will be be due the beginning of class on the due date, unless otherwise specified. Otherwise, a homework will be worth 50% if one day late, and 0% if it is two or more days late. It is required to submit all homeworks even if after two days, if you do not want an incomplete grade. The last last one-third of class on readings of applications of graphical models. I will provide a list of candidate applications, along with some papers for each application, but feel free to supplement these. Teams -- of size 3 or 4 -- each picks one application; and will provide a brief presentation of their findings in class; which I will add to.
Class Particip. policy This is a graduate class; I expect students to participate actively in the class.