Computing That Serves

Using Structured Representations in Nonparametric Bayesian Models


Thursday, April 3, 2008 - 11:00am


Tom Griffiths, Assistant Professor of Psychology and Cognitive Science, UC Berkeley

Methods from nonparametric Bayesian statistics are often used to address model selection problems such as determining the number of components in a mixture model. At the heart of these methods is the Chinese restaurant process (CRP), which defines a distribution on assignments of observations to components in a way that does not limit the number of components. In this work, we expand the class of statistical models to which nonparametric Bayesian methods can be applied by defining distributions with similar properties on two kinds of structured representations: trees and binary matrices. These distributions can be used as priors in models used in a variety of applications -- anywhere data is to be explained in terms of a latent hierarchy, set of features, or bipartite graph, but the dimensionality of this latent structure is unknown. I will describe some applications to modeling text, images, and psychological data. This is joint work with David Blei, Zoubin Ghahramani, and Michael Jordan.


Tom Griffiths is an Assistant Professor of Psychology and Cognitive Science at UC Berkeley, with courtesy appointments in Computer Science and Neuroscience. His research explores connections between human and machine learning, using ideas from statistics and artificial intelligence to try to understand how people solve the challenging computational problems they encounter in everyday life. He received his PhD in Psychology from Stanford University in 2005, and taught in the Department of Cognitive and Linguistic Sciences at Brown University before moving to Berkeley. His work and that of his students has received awards from the Neural Information Processing Systems conference and the Annual Conference of the Cognitive Science Society, and in 2006 IEEE Intelligent Systems magazine named him one of "Ten to watch in AI."