Non-Reversible Parallel Tempering: a Scalable Highly Parallel MCMC Scheme

Abstract

Parallel tempering (PT) methods are a popular class of Markov chain Monte Carlo schemes used to sample complex high-dimensional probability distributions. They rely on a collection of $N$ interacting auxiliary chains targeting tempered versions of the target distribution to improve the exploration of the state-space. We provide here a new perspective on these highly parallel algorithms and their tuning by identifying and formalizing a sharp divide in the behaviour and performance of reversible versus non-reversible PT schemes. We show theoretically and empirically that a class of non-reversible PT methods dominates its reversible counterparts and identify distinct scaling limits for the non-reversible and reversible schemes, the former being a piecewise-deterministic Markov process and the latter a diffusion. These results are exploited to identify the optimal annealing schedule for non-reversible PT and to develop an iterative scheme approximating this schedule. We provide a wide range of numerical examples supporting our theoretical and methodological contributions. The proposed methodology is applicable to sample from a distribution $\pi$ with a density $L$ with respect to a reference distribution ${\pi }_0$ and compute the normalizing constant. A typical use case is when ${\pi }_0$ is a prior distribution, $L$ a likelihood function and $\pi$ the corresponding posterior.

74 pages, 30 figures. The method is implemented in an open source probabilistic programming available at https://github.com/UBC-Stat-ML/blangSDK

References in corpus (7)

• First M87 Event Horizon Telescope Results. VII. Polarization of the Ring
• Feedback-optimized parallel tempering Monte Carlo
• Where Are We With Light Sterile Neutrinos?
• Conditions for rapid mixing of parallel and simulated tempering on multimodal distributions
• On Dynamics and Optimal Number of Replicas in Parallel Tempering Simulations
• Persistent Non-Gaussian Structure in the Image of Sagittarius A* at 86 GHz
• Blang: Bayesian declarative modelling of general data structures and inference via algorithms based on distribution continua

Cited by in corpus (20)

• First M87 Event Horizon Telescope Results. VII. Polarization of the Ring
• First Sagittarius A* Event Horizon Telescope Results. IV. Variability, Morphology, and Black Hole Mass
• Adaptive tempered reversible jump algorithm for Bayesian curve fitting
• The cool brown dwarf Gliese 229 B is a close binary
• Efficient Irreversible Monte Carlo samplers
• Learning Regionalization using Accurate Spatial Cost Gradients within a Differentiable High-Resolution Hydrological Model: Application to the French Mediterranean Region
• Bayesian Black Hole Photogrammetry
• Accelerating equilibrium spin-glass simulations using quantum annealers via generative deep learning
• Keplerian motion of a compact source orbiting the inner disc of PDS 70: a third protoplanet in resonance with b and c?
• Message Passing Variational Autoregressive Network for Solving Intractable Ising Models
• An asymptotic Peskun ordering and its application to lifted samplers
• Origin of the ring ellipticity in the black hole images of M87*
• Parallel Tempering on Optimized Paths
• All in the Exponential Family: Bregman Duality in Thermodynamic Variational Inference
• Hamiltonian Assisted Metropolis Sampling
• Quantum-Inspired Tempering for Ground State Approximation using Artificial Neural Networks
• A connection between Tempering and Entropic Mirror Descent
• Sampling metastable systems using collective variables and Jarzynski-Crooks paths
• An Embedded Diachronic Sense Change Model with a Case Study from Ancient Greek
• Theoretical guarantees for lifted samplers