Title: Nonnegative Bayesian nonparametric factor models with completely random measures
Authors: Fadhel Ayed - Oxford (United Kingdom) [presenting]
Francois Caron - University of Oxford (United Kingdom)
Abstract: Bayesian nonparametric Poisson factor models for community detection are studied. We assume that each user is affiliated to an infinite number of communities, which are modeled using a Completely Random Measure (CRM). The model is flexible and allows the number of active communities to grow unboundedly with the number of users. We show how the properties of the CRM relate to the growth of the number of active communities. We also derive a Markov chain Monte Carlo algorithm for posterior inference for this class of models.