Abstract:Parallel Markov Chain Monte Carlo (pMCMC) algorithms generate clouds of proposals at each step to efficiently resolve a target probability distribution. We build a rigorous foundational framework for pMCMC algorithms that situates these methods within a unified 'extended phase space' measure-theoretic formalism. Drawing on our recent work that provides a comprehensive theory for reversible single proposal methods, we herein derive general criteria for multiproposal acceptance mechanisms which yield ergodic chains on general state spaces. Our formulation encompasses a variety of methodologies, including proposal cloud resampling and Hamiltonian methods, while providing a basis for the derivation of novel algorithms. In particular, we obtain a top-down picture for a class of methods arising from 'conditionally independent' proposal structures. As an immediate application, we identify several new algorithms including a multiproposal version of the popular preconditioned Crank-Nicolson (pCN) sampler suitable for high- and infinite-dimensional target measures which are absolutely continuous with respect to a Gaussian base measure. To supplement our theoretical results, we carry out a selection of numerical case studies that evaluate the efficacy of these novel algorithms. First, noting that the true potential of pMCMC algorithms arises from their natural parallelizability, we provide a limited parallelization study using TensorFlow and a graphics processing unit to scale pMCMC algorithms that leverage as many as 100k proposals at each step. Second, we use our multiproposal pCN algorithm (mpCN) to resolve a selection of problems in Bayesian statistical inversion for partial differential equations motivated by fluid measurement. These examples provide preliminary evidence of the efficacy of mpCN for high-dimensional target distributions featuring complex geometries and multimodal structures.

Within‐chain parallelization—Giving Stan Jet Fuel for population modeling in pharmacometrics

Flexible and efficient Bayesian pharmacometrics modeling using Stan and Torsten, Part I

Adaptable Parallel Acceleration Strategy For Dynamic Monte Carlo Simulations Of Polymerization With Microscopic Resolution

Parallelization and High-Performance Computing Enables Automated Statistical Inference of Multi-scale Models.

Accelerating sequential Monte Carlo with surrogate likelihoods

Parallelization of Kinetic Theory Simulations

SMC Is All You Need: Parallel Strong Scaling

Parallel Markov Chain Monte Carlo for Non-Gaussian Posterior Distributions

A Markov Chain Monte-Carlo Approach to Dose-Response Optimization Using Probabilistic Programming (RStan)

Single MCMC Chain Parallelisation on Decision Trees

Hamiltonian Monte Carlo Acceleration Using Surrogate Functions with Random Bases

Parallel and interacting Markov chains Monte Carlo method

Parallel MCMC Algorithms: Theoretical Foundations, Algorithm Design, Case Studies

Orthogonal parallel MCMC methods for sampling and optimization

Vector operations for accelerating expensive Bayesian computations -- a tutorial guide

Two General Methods for Population Pharmacokinetic Modeling: Non-Parametric Adaptive Grid and Non-Parametric Bayesian

Accelerating inference for stochastic kinetic models

Accelerating delayed-acceptance Markov chain Monte Carlo algorithms

Parallel Tempering for Dynamic Generalized Linear Models

ParaDRAM: A Cross-Language Toolbox for Parallel High-Performance Delayed-Rejection Adaptive Metropolis Markov Chain Monte Carlo Simulations

Accelerating Markov Chain Monte Carlo sampling with diffusion models