Exploring Discrete Analogs of Hamiltonian Monte Carlo

09/11/2019
by   Guangyao Zhou, et al.
0

Hamiltonian Monte Carlo (HMC) has emerged as a powerful way to sample from continuous distributions. However, a fundamental limitation of HMC is that it can't be applied to discrete distributions. In this paper, we propose the momentum sampler as a general framework for exploring the idea of building discrete analogs of HMC. The momentum sampler is a novel family of Markov Chain Monte Carlo (MCMC) algorithms that extends arbitrary single-site Metropolis Hastings (MH) samplers for discrete distributions with an auxiliary momentum process. It connects and generalizes a variety of existing methods, and points to a more general framework where we can mix stochastic evolution of discrete variables with deterministic evolution of continuous variables. The auxiliary momentum process can be integrated exactly, resulting in a rejection-free sampling process. The various trade-offs involved in picking different proposal distributions and kinetic energies have clear physical interpretations, and are demonstrated by numerical experiments on a simple Potts model.

READ FULL TEXT
research
02/12/2018

Bouncy Hybrid Sampler as a Unifying Device

This work introduces a class of rejection-free Markov chain Monte Carlo ...
research
10/21/2021

Hamiltonian Monte Carlo with Asymmetrical Momentum Distributions

Existing rigorous convergence guarantees for the Hamiltonian Monte Carlo...
research
04/19/2020

Zig-zag sampling for discrete structures and non-reversible phylogenetic MCMC

We construct a zig-zag process targeting posterior distributions arising...
research
01/31/2020

Non-reversibly updating a uniform [0,1] value for Metropolis accept/reject decisions

I show how it can be beneficial to express Metropolis accept/reject deci...
research
12/15/2021

The Apogee to Apogee Path Sampler

Amongst Markov chain Monte Carlo algorithms, Hamiltonian Monte Carlo (HM...
research
03/01/2020

Stein Variational Inference for Discrete Distributions

Gradient-based approximate inference methods, such as Stein variational ...

Please sign up or login with your details

Forgot password? Click here to reset