Markov chain Monte Carlo without evaluating the target: an auxiliary variable approach

TL;DR

This paper introduces a new auxiliary variable framework for MCMC that enables sampling without evaluating the target distribution, improving efficiency especially for complex or intractable models.

Contribution

It unifies existing MCMC algorithms under a common framework and develops novel algorithms using auxiliary variables and estimated gradients for better performance.

Findings

New algorithms outperform existing methods on synthetic data

Framework simplifies and extends previous MCMC results

Demonstrates effectiveness on real datasets

Abstract

In sampling tasks, it is common for target distributions to be known up to a normalizing constant. However, in many situations, even evaluating the unnormalized distribution can be costly or infeasible. This issue arises in scenarios such as sampling from the Bayesian posterior for tall datasets and the 'doubly-intractable' distributions. In this paper, we begin by observing that seemingly different Markov chain Monte Carlo (MCMC) algorithms, such as the exchange algorithm, PoissonMH, and TunaMH, can be unified under a simple common procedure. We then extend this procedure into a novel framework that allows the use of auxiliary variables in both the proposal and the acceptance-rejection step. Several new MCMC algorithms emerge from this framework that utilize estimated gradients to guide the proposal moves. They have demonstrated significantly better performance than existing methods on both synthetic and real datasets. Additionally, we develop the theory of the new framework and apply it to existing algorithms to simplify and extend their results.