Dependent Probabilistic Matrix Factorization

Sunday, March 14, 2010
7:00 PM
Free and open to the public

Probabilistic matrix factorization (PMF) is a powerful method for modeling data associated with pairwise relationships. PMF is used in collaborative filtering, computational biology, and document analysis among other application areas. In many of these domains, additional information is available that might assist in the modeling of the pairwise interactions. For example, when modeling movie ratings, we might know when the rating occurred, where the user lives, or what actors appear in the movie. It has been difficult, however, to incorporate this kind of side information into the probabilistic matrix factorization model. I will discuss some recent work to develop a nonparametric Bayesian framework for incorporating side information by coupling together multiple PMF problems via Gaussian process priors. I will show how we have used this model to successfully model the outcomes of NBA basketball games, allowing for the attributes of the teams to vary over time and to include home-team advantage. This is joint work with George Dahl and Iain Murray.

x x


Ryan Adams

CIFAR Junior Research Fellow
University of Toronto