Seminars/Colloquia/Invited Talks

Seminars

Jonathan Yedidia

Understanding Belief Propagation

PLACE:	Clark 314
EVENT:	CIS Seminar Series
DATE:	April 1, 2005
TIME:	1:00 - 2:00

Abstract

Belief propagation (BP) algorithms can be applied to a wide range of "inference" problems, including those that arise in communications, statistical physics, artificial intelligence, and image processing. I will first explain how these problems are formulated using "factor graphs," and then try to give an intuitive explanation for BP algorithms that operate by sending messages between the nodes of the factor graphs. I will also explain the close relationship between BP algorithms and "free energy" approximations from statistical physics. Understanding this relationship makes possible the development of improved "generalized" belief propagation algorithms.

Brief biography

Jonathan Yedidia is a Research Scientist at Mitsubishi Electric Research Laboratories (MERL), where he has been since 1998. His main research interests are in probabilistic inference algorithms, and their applications in coding theory and signal processing. He received his Ph.D. in theoretical physics from Princeton University in 1990.