Unbiased Markov Chain Monte Carlo: what, why, and how

TL;DR

This paper introduces methods to eliminate initialization bias in MCMC estimates, emphasizing practical methodology, with implications for parallel computing and diagnostics, while briefly touching on theoretical aspects.

Contribution

It provides an accessible introduction to bias removal techniques in MCMC, focusing on practical methods rather than deep theoretical analysis.

Findings

Methods effectively remove burn-in bias from MCMC estimates.

Implications for parallel computing and convergence diagnostics are discussed.

Focus on practical methodology over theoretical details.

Abstract

This document presents methods to remove the initialization or burn-in bias from Markov chain Monte Carlo (MCMC) estimates, with consequences on parallel computing, convergence diagnostics and performance assessment. The document is written as an introduction to these methods for MCMC users. Some theoretical results are mentioned, but the focus is on the methodology.