Modified Hamiltonian Monte Carlo for Bayesian inference

Abstract

The Hamiltonian Monte Carlo (HMC) method has been recognized as a powerful sampling tool in computational statistics. We show that performance of HMC can be significantly improved by incorporating importance sampling and an irreversible part of the dynamics into a chain. This is achieved by replacing Hamiltonians in the Metropolis test with modified Hamiltonians, and a complete momentum update with a partial momentum refreshment. We call the resulting generalized HMC importance sampler---Mix & Match Hamiltonian Monte Carlo (MMHMC). The method is irreversible by construction and further benefits from (i) the efficient algorithms for computation of modified Hamiltonians; (ii) the implicit momentum update procedure and (iii) the multi-stage splitting integrators specially derived for the methods sampling with modified Hamiltonians. MMHMC has been implemented, tested on the popular statistical models and compared in sampling efficiency with HMC, Riemann Manifold Hamiltonian Monte Carlo, Generalized Hybrid Monte Carlo, Generalized Shadow Hybrid Monte Carlo, Metropolis Adjusted Langevin Algorithm and Random Walk Metropolis-Hastings. To make a fair comparison, we propose a metric that accounts for correlations among samples and weights, and can be readily used for all methods which generate such samples. The experiments reveal the superiority of MMHMC over popular sampling techniques, especially in solving high dimensional problems.

arXiv comment30 pages, 17 figures

References in corpus (39)

• A Conceptual Introduction to Hamiltonian Monte Carlo
• Stochastic Gradient Hamiltonian Monte Carlo
• Variance Reduction using Nonreversible Langevin Samplers
• Annealed Importance Sampling
• On the Geometric Ergodicity of Hamiltonian Monte Carlo
• Importance Tempering
• Numerical integrators for the Hybrid Monte Carlo method
• Cost of the Generalised Hybrid Monte Carlo Algorithm for Free Field Theory
• Compressible Generalized Hybrid Monte Carlo
• Improving Asymptotic Variance of MCMC Estimators: Non-reversible Chains are Better
• A General Metric for Riemannian Manifold Hamiltonian Monte Carlo
• Spherical Hamiltonian Monte Carlo for Constrained Target Distributions
• Hamiltonian Monte Carlo Without Detailed Balance
• Auxiliary-variable Exact Hamiltonian Monte Carlo Samplers for Binary Distributions
• Weighted skewness and kurtosis unbiased by sample size
• Skewness and kurtosis unbiased by Gaussian uncertainties
• Lagrangian Dynamical Monte Carlo
• Generalizing the No-U-Turn Sampler to Riemannian Manifolds
• Extra Chance Generalized Hybrid Monte Carlo
• Probabilistic Path Hamiltonian Monte Carlo
• Gradient-free Hamiltonian Monte Carlo with Efficient Kernel Exponential Families
• Adaptive Hamiltonian and Riemann Manifold Monte Carlo Samplers
• Nested Sampling with Constrained Hamiltonian Monte Carlo
• Hamiltonian Annealed Importance Sampling for partition function estimation
• Magnetic Hamiltonian Monte Carlo
• Mathematical understanding of detailed balance condition violation and its application to Langevin dynamics
• Multi-stage splitting integrators for sampling with modified Hamiltonian Monte Carlo methods
• Nonreversible Langevin Samplers: Splitting Schemes, Analysis and Implementation
• Towards Unifying Hamiltonian Monte Carlo and Slice Sampling
• Adiabatic Monte Carlo
• Geometrically Tempered Hamiltonian Monte Carlo
• Semi-Separable Hamiltonian Monte Carlo for Inference in Bayesian Hierarchical Models
• Roll-back Hamiltonian Monte Carlo
• Hamiltonian Monte Carlo with Reduced Momentum Flips
• Control of probability flow in Markov chain Monte Carlo -- Nonreversibility and lifting
• Stochastic Gradient Monomial Gamma Sampler
• Stochastic Variance-Reduced Hamilton Monte Carlo Methods
• Tempering by Subsampling
• Continuously tempered Hamiltonian Monte Carlo

Cited by in corpus (9)

• Post-Processing of MCMC
• Multi-stage splitting integrators for sampling with modified Hamiltonian Monte Carlo methods
• Geometric Methods for Sampling, Optimisation, Inference and Adaptive Agents
• Sampling algorithms in statistical physics: a guide for statistics and machine learning
• Adaptive multi-stage integration schemes for Hamiltonian Monte Carlo
• A Neural Network MCMC sampler that maximizes Proposal Entropy
• HMC: avoiding rejections by not using leapfrog and some results on the acceptance rate
• A Unifying and Canonical Description of Measure-Preserving Diffusions
• Fast generation of Gaussian random fields for direct numerical simulations of stochastic transport