Scalable Multiscale Bayesian Models

• Activity Number: 310
• Type: Invited
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): David Dunson*+
• Companies: Duke University
• Keywords: Bayesian ; nonparametrics ; multiscale ; big data ; MCMC ; approximation

Abstract

Analyses of big data have been dominated in recent years by penalized optimization methods, with Lasso and its variants taking center stage. In many big data settings, ranging from image processing to nonparametric regression, multiscale methods have proven to provide state of the art performance. We are motivated to move beyond point estimation to provide a probabilistic characterization of uncertainty using Bayesian methods. This talk will focus on scaling up Bayesian nonparametric inferences to big data problems through the use of multiscale methods combined with novel algorithms. We are particularly interested in novel adaptations of MCMC algorithms to allow dramatically faster and more robust implementations including in huge data and richly parameterized models.