A1107
Title: Parallel computing methods for Bayesian analysis of big data sets
Authors: Erin Conlon - University of Massachusetts Amherst (United States) [presenting]
Zheng David Wei - Texas A&M University - Corpus Christi (United States)
Abstract: Recently, new parallel Bayesian Markov Chain Monte Carlo (MCMC) methods have been developed for massive data sets that are too large for traditional statistical analysis. These methods partition big data sets by observations into subsets. The purpose is to discuss the alternative parallel Bayesian MCMC computing algorithm that partitions big data sets by groups rather than observations. This two-stage approach analyzes groups independently in parallel in stage one; the posteriors from stage one are used as proposal distributions in stage two, which estimates the complete data model. The method is illustrated with both three-level and four-level models, and improvements are shown in computation time as well as MCMC efficiency versus the complete data evaluation.
