Evaluating Scalable Markov Chain Monte Carlo Algorithms for Big Data Problems Open Access

Abstract

Advances in technology have led to generation of enormous amounts of data in many fields including medicine, presenting challenges in data analysis. The reason could be processor, memory, or disk storage bottlenecks in computational environments. These challenges are particularly pronounced for Bayesian analysis, which often entails the use of Markov chain Monte Carlo (MCMC) in computation. As such, a number of scalable MCMC algorithms have been developed to alleviate the computational challenges in three main directions. The first direction is to accelerate expensive gradient computation at each MCMC iteration. The second direction is to parallelize computation at each MCMC iteration, requiring potentially expensive communication within each iteration. The third direction is to divide a data set into small subsets and run independent MCMC for each subset before combining them. This work focuses on scalable MCMC algorithms developed in the third direction. We conduct simulation studies to evaluate and compare several parallel MCMC algorithms. Examples of scalable MCMC are shown for Bayesian linear regression and other regressions.

Table of Contents

1. Introduction

1.1 Background

1.2 Current Approaches

1.2.1 Nonparallel Accelerating Approach

1.2.2 Communication-Intense Parallel Approach

1.2.3 Communication-Free Parallel Approach

1.3 General Comparisons

2. Analysis of MCMC Algorithms and Computational Challenges

2.1 Overview

2.2 Algorithms in Nonparallel Accelerating Approach

2.3 Communication-Intense Parallel Algorithms

2.4 Communication-Free Parallel Algorithms

3. Simulation Study

3.1 Goals

3.2 Simulation Settings

3.3 Performance Metrics

3.4 Results

4. Discussion

References

About this Master's Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Rollins School of Public Health
Department	Biostatistics and Bioinformatics
Subfield / Discipline	Biostatistics - MPH & MSPH
Degree	MSPH
Submission	Master's Thesis
Language	English
Research Field	Biology, Biostatistics
Keyword	MCMC Bayesian Inference Parallel Computing
Committee Chair / Thesis Advisor	Long, Qi, Emory University
Committee Members	Kundu, Suprateek , Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Evaluating Scalable Markov Chain Monte Carlo Algorithms for Big Data Problems ()	2018-08-28 10:09:03 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions