Events - Colloquia & Seminars

CCIS Colloquium Spring 2007

Probabilistic Models with Unknown Objects

Speaker: Brian Milch

Affiliation: MIT

Date: Wednesday, January 10, 2007

Talk: 12:00 pm, 366 WVH

Abstract

Many AI problems, from tracking aircraft based on radar blips to extracting facts about people and events from text documents, involve making inferences about the real-world objects that underlie some data. In many cases, we do not know the number of underlying objects or the mapping between objects and observations. This talk will present a probabilistic modeling language, called Bayesian logic (or BLOG), which allows us to represent such scenarios in a natural way. A well-formed BLOG model fully defines a probability distribution over model structures of a first-order logical language; these "possible worlds" can contain varying numbers of objects with varying relations among them. I will also describe a Markov chain Monte Carlo algorithm for performing inference on BLOG models. This algorithm is novel in that it does a random walk not over fully specified possible worlds, but over partial world descriptions that instantiate only the relevant variables. I will present the results of applying this algorithm to identify the distinct publications referred to by a set of citation strings extracted from online papers.

The recent paper First-Order Probabilistic Languages: Into the Unknown gives a survey of first-order probabilistic languages.

Biography

Brian Milch is a postdoctoral researcher in the Computer Science and Artificial Intelligence Laboratory at MIT. He received his B.S. with honors in Symbolic Systems from Stanford University, where he worked with Prof. Daphne Koller. He then spent a year as a research engineer at Google before entering the computer science Ph.D. program at U.C. Berkeley. His thesis research, with Prof. Stuart Russell, was on representation and inference for models that combine probability and first-order logic. He received his Ph.D. in December 2006. He is also the recipient of an NSF Graduate Research Fellowship and a Siebel Scholarship.