Speaker: Ryan Prescott Adams
Electrical and Computer Engineering Department
University of Toronto
Title: Infinite Belief Networks
Abstract: Belief networks are a powerful representation for modeling complex probability distributions. Learning the structure of a belief network, particularly one with hidden units, has been difficult, however. The Indian buffet process (IBP) has previously been used as a nonparametric Bayesian prior on the directed structure of a belief network with a single infinitely wide hidden layer. In this talk, I will describe the cascading Indian buffet process (CIBP), which provides a nonparametric Bayesian prior on the structure of a layered, directed belief network that is both infinitely deep and infinitely wide. The CIBP results in networks in which a provably finite yet unbounded set of hidden units contribute to the distribution over visible units. I will discuss using the CIBP prior with the continuous sigmoidal belief network so that not only is there flexibility in the number of hidden units and their edge structure, but also each unit can vary its behavior between discrete and continuous states. I will also discuss Markov chain Monte Carlo algorithms for inference in these belief networks. This is joint work with Hanna Wallach (UMass-Amherst) and Zoubin Ghahramani (Cambridge Engineering and CMU).
For Additional Information contact:
Hugo Larochelle http://www.cs.toronto.edu/~larocheh/