Machine Learning Meetings and Events

Group Meetings: Group meetings are held Mondays from 11am to Noon (talk starts 11:10am) in D.L. Pratt 290C unless otherwise noted. Meetings are coordinated by Hugo Larochelle.

Tea Talks: Tea talks are held every Wednesday at 4:00pm in D.L. Pratt 290C. Talks should be simple, accessible, and not exceed 15 minutes. Speakers bring snacks, make tea, and provide a copy of the presented paper.

Tea Talk Mar 4, 2009: Representing structured knowledge using hierarchical Bayesian models.

Speaker: Ilya Sutskever
Abstract: I will present work on representing structured knowledge using hierarchical Bayesian models. More specifically, I will talk about: 1) Ways of defining and learning complicated prior distributions over attributes of items. This distribution, when learned, can be used to evaluate the strength of statements of the form "if cows have X and horses have X, do all mammals have X?". 2) A prior over prior distributions over graphs that are used as structured priors for directed models. It can be used to infer which variables cause which variables from a small number of examples, by learning that there exist classes of objects that influence other classes of objects in a "uniform" way. In both cases, a discrete structure (such as a tree or a grammar) is used to define the prior distribution over the objects of interest (the object's attributes and graphs). The posterior over the discrete structures is the knowledge of the system. In both cases the model is probabilistic, so it works well even when the data is noisy and does not completely satisfy the model's assumption, unlike much earlier word on structured knowledge. References: Structured statistical models of inductive reasoning. Kemp, C. and Tenenbaum, J. B http://web.mit.edu/cocosci/Papers/struct-stat-inpress.pdf Intuitive theories as grammars for causal inference. Tenenbaum, J.B., Griffiths, T. L., and Niyogi, S. http://web.mit.edu/cocosci/Papers/tgn-grammar.pdf