Expectation Maximization for Distributed Computing

• Activity Number: 458
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Glen DePalma*+ and Sanvesh Srivastava and Chuanhai Liu
• Companies: Purdue University and Purdue University and Purdue University
• Keywords: EM ; Distributed Computing ; Parallel Computing ; Computational ; R

Abstract

The Expectation-Maximization (EM) algorithm (Dempster, Laird, and Rubin, 1977) remains a hallmark achievement in the history of Statistics and optimization. Most of the research in the implementation of EM algorithm addresses the fundamental issue of making the EM algorithm efficient by speeding up its rate of convergence. Despite the widespread attention and substantial effort over the past 30+ years, however, there has been no systematic attempt to take advantage of distributed computing. This issue is related to the underlying iterative nature of the algorithm that makes EM difficult to parallelize. We extend the basic EM algorithm to a parallel framework that takes advantage of cluster computing and multiprocessing environments in computer architecture. Furthermore, we provide an R (R Development Core Team, 2012) implementation based on the developmental DISC package that facilitates the easy use of the parallel-EM framework.