Faster parallel MCMC: Metropolis adjustment is best served warm

TL;DR

This paper introduces LAPS, a parallel MCMC method that uses an unadjusted phase to quickly approach the target distribution before applying Metropolis adjustment, significantly speeding up sampling.

Contribution

The paper presents LAPS, a practical, hyperparameter-free scheme that combines unadjusted and adjusted MCMC phases for faster parallel sampling.

Findings

LAPS outperforms ensemble adjusted methods like MEADS and ChESS.

LAPS achieves two orders of magnitude faster sampling than sequential algorithms like NUTS.

LAPS automatically selects hyperparameters without manual tuning.

Abstract

Despite the enormous success of Hamiltonian Monte Carlo and related Markov Chain Monte Carlo (MCMC) methods, sampling often still represents the computational bottleneck in scientific applications. Availability of parallel resources can significantly speed up MCMC inference by running a large number of chains in parallel, each collecting a single sample. However, the parallel approach converges slowly if the chains are not initialized close to the target distribution (cold start). Theoretically this can be resolved by initially running MCMC without Metropolis-Hastings adjustment to quickly converge to the vicinity of the target distribution and then turn on adjustment to achieve fine convergence. However, no practical scheme uses this strategy, due to the difficulty of automatically selecting the step size during the unadjusted phase. We here develop Late Adjusted Parallel Sampler (LAPS), which is precisely such a scheme and is applicable out of the box, all the hyperparameters are selected automatically. LAPS takes advantage of ensemble-based hyperparameter adaptation to estimate the bias at each iteration and converts it to the appropriate step size. We show that LAPS consistently and significantly outperforms ensemble adjusted methods such as MEADS or ChESS and the optimization-based initializer Pathfinder on a variety of standard benchmark problems. LAPS typically achieves two orders of magnitude lower wall-clock time than the corresponding sequential algorithms such as NUTS.