Time Schedule:
Jeffrey A. Bilmes
E E 596
Seattle Campus
Topics of current interest in signal and image processing. Content may vary from offering to offering. Prerequisite: permission of instructor.
Class description
Course Announcement Pattern recognition from a Graphical Models perspective EE596B Statistical Pattern Recognition II
Spring 2004 Instructor: Jeff A. Bilmes, 418 EE/CS Bldg. Email: bilmes@ee.washington.edu
Course web page: http://ssli.ee.washington.edu/courses/PRII Units: 4 Course Hours: TTh 3:30-5:20 EE1-037 ***NOTE***: First meeting will be Thursday, April 1st at 4:30pm. Due to seminar conflicts, the course will occasionally run from 4:30-6:20pm.
Description: This course will serve as a general introduction to graphical models and Bayesian networks for pattern recognition and statistical learning. Topics will include:
1) Graph semantics: conditional independence; semantics of directed and undirected models; algorithms for conditional independence including Bayes-ball, d-separation, Markov properties on graphs, factorization.
2) Models: linear Gaussian models, mixture models, factor analysis, probabilistic decision trees, Markov Random Fields, Gibbs distributions, conditional random fields, multivariate Gaussians as graphical models.
3) Dynamic (temporal) models: Hidden Markov Models, Kalman filtering and linear-Gaussian HMMs, dynamic Bayesian networks (DBNs)
4) Graph Theory: moralization; triangulated, decomposable, and intersection graphs, k-trees, hypergraphs.
5) Probabilistic Inference: junction trees, belief propagation and its guises (including Pearl formulation, Hugin, Shafer-Shenoy, Bucket-elimination, etc.); join-trees and data-strucure representations thereof; methods to optimally triangulate graphs; generalizations including mean-field and variational, loopy-propagation, and sampling.
6) Learning: learning Bayesian networks, EM for parameter and structure learning, alternating minimization, discriminative parameter and structure learning, other learning methods.
7) Models in practice: QMR and Factorial HMMs, turbo-coding and other codes on graphs (and the belief propagation algorithm), various dynamic models, etc. Practical issues of using graphical models and Bayesian networks for real-world problems.
We will also briefly introduce kernel machines, and topics such as bootstrapping, model averaging, bagging, and boosting.
The course will have homework assignments, a midterm exam, and a final project (a research paper and final presentation). There will be no final exam.
Course materials: Main Texts: 1) "An Introduction to Graphical Models", by Mike Jordan (we will use a pre-published version directly from the authors) 2) Online course reader
Other Texts:
1) Chapters from "Graphical Models" by Steffen Lauritzen, Oxford Science Publications
Reference Texts: 1) "Neural Networks for Pattern Recognition", by C. Bishop 2) "The Elements of Statistical Learning: Data Mining, Inference," and Prediction Hastie, Tibshirani, and Friedman 3) "Pattern Classification," R. Duda, P. Hart and D. Stork
Prerequisite: Statistical Pattern Recognition I, or prior exposure to pattern recognition concepts, or consent of instructor.
Student learning goals
General method of instruction
Recommended preparation
Prerequisites: basic probability, statistics, and random processes Basic knowledge of matlab. The course is open to students in all UW departments.
Class assignments and grading
