Independent Approximate Draws from High-Dimensional Intractable Probability Distributions

• Activity Number: 376
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Andrew Olsen*+ and Radu Herbei
• Companies: The Ohio State University and The Ohio State University
• Keywords: Exact Sampling ; MCMC ; High-dimensional

Abstract

Obtaining independent exact draws from an intractable probability distribution is often challenging and frequently impossible, especially as the dimension of the distribution increases. We propose a method that produces independent draws which are approximately distributed according to the target distribution. This method is applicable in many settings, including high dimensions. One advantage of this method over traditional Markov chain Monte Carlo methods is the ability to obtain the draws in parallel, which can lead to large computational savings. We illustrate this method with examples.