Michael Jordan Pehong Chen Distinguished Professor
Department of Electrical Engineering Computer Science
Department of Statistics
University of California, Berkeley

"Nonparametric Graphical Models"

Graphical models have classically been developed within the realm of parametric statistics---the distributions that have been considered are the multinomial and the Gaussian and (occasionally) other distributions in the exponential family, and the set of graphs under consideration in a given problem is generally taken to be fixed and finite.  In this talk I overview some of the stochastic processes that are allowing us to move beyond these classical parametric restrictions; these include the Dirichlet process, the beta process, the gamma process, and various recursive constructions (e.g., Bayesian hierarchies) involving these processes.  I also discuss some of the marginals that arise from integrating out these stochastic processes; these are guaranteed to enhance your appetite for nonparametrics.

Thursday, April 17th, 2008
11:20am -  12:00 pm
IRCS

3401 Walnut Street, Room 400A

For more information regarding our speaker please visit:

http://www.cs.berkeley.edu/~jordan/