Stat /Biostat 538: Scientific and Statistical Computing

Winter 2009

Instructor: Thomas Richardson
Office: Padelford B303
Email: thomasr [at] u [dot] washington [dot] edu
Office Hours: After Class or e-mail me to arrange a time.
When: T,Th 11:30 - 1:00
Where: Communications (CMU) 228

Overview

The theme of the present course will be on graphical models, which are multivariate statistical models. The models are graphical in the sense that they may be represented by a graph in which each node corresponds to a random variable, and the missing edges between the nodes represent conditional independencies. 'Bayesian' networks are a well-known example of a graphical model. We will examine the algorithmic, computational and practical aspects of these models. A particular emphasis of the course will be optimization methods used for parameter estimation in graphical models. We will also examine other topics including methods for storing and manipulating probability distributions represented by these models. In addition we will look at techniques for performing model selection.

Intended Audience

This class is intended primarily for graduate students with an interest in statistics, algorithms and computing.

Evaluation

The grade will be based on a project (40%), homework (50%) and class participation (10%). Students taking the course for credit will complete a programming project (possibly in pairs) extending through most of the quarter. Each pair will create an outline project plan in the first couple of weeks of the quarter. Students are encouraged to choose a project topic related to their own research interests, but should discuss the topic with the instructor before embarking upon it.

Pre-requisites:

Either STAT 534 OR If you intend to take this class, please send me e-mail, so that I can get a rough estimate of the class size.