A0428
Title: Consistency of variational approximations in misspecified models
Authors: Pierre Alquier - ESSEC Business School (Singapore) [presenting]
Abstract: Variational Bayesian algorithms (VB) aims at approximating the posterior by a distribution in a tractable family. Thus, MCMC are replaced by an optimization algorithm which is orders of magnitude faster. VB methods have been applied in such computationally demanding applications as including collaborative filtering, image processing, NLP and text processing... However, despite nice results in practice, the theoretical properties of these approximations were not known until the past two years. We will review some of these recent results. We will emphasize that approximations of tempered posteriors are more robust to model misspecification. We will also discuss efficient optimization procedures for robust-VB.
