High-Dimensional Nonparametric Bayes Modeling via Nonlinear Latent Factor Models

• Activity Number: 497
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): David Dunson*+
• Companies: Duke University
• Keywords: Gaussian process ; factor analysis ; nonparametric Bayes ; dimensionality reduction

Abstract

It has become routine to encounter massive dimensional data in a broad variety of application areas. One useful tool for dimensionality reduction and modeling is sparse Bayesian latent factor analysis. This talk focuses on developing general frameworks for making such approaches more flexible, and hence facilitate a greater degree of dimensionality reduction as well as nonparametric density estimation and regression in massive dimensions. A class of infinite latent factor models is proposed, with the higher indexed factors having decreasing impact on the response, while allowing nonlinearities. Theoretical properties are considered and the methods are applied to simulated data and applications in genomics and machine learning.