Speeding Up Nonsmooth Bayesian MCMC Sampling via Inexact Proximal Unadjusted Langevin Algorithm

TL;DR

This paper introduces iPULA, an inexact proximal Langevin algorithm for efficient sampling from nonsmooth posterior distributions, with theoretical guarantees and practical validation on medical imaging.

Contribution

The paper proposes iPULA, a novel inexact proximal Langevin method that handles nonsmooth potentials with controlled approximation errors, extending practical applicability.

Findings

iPULA converges with quantifiable bias due to inexactness.

Theoretical guarantees relate inexactness to sampling error.

Experimental results on medical imaging validate effectiveness.

Abstract

We study sampling from posterior distributions with nonsmooth composite potentials, a setting in which proximal-based Langevin methods are theoretically appealing but in practice limited to simple functions with closed-form proximal operators. We introduce iPULA for composite potentials, an inexact proximal unadjusted Langevin algorithm that replaces exact proximal steps with controlled approximations. Our approach leverages the Moreau envelope to smooth the potential, while allowing inexact evaluation of its gradient through inexact proximal computations. We establish non-asymptotic convergence guarantees for iPULA, explicitly characterizing the impact of inexactness on the sampling error and showing that the inexactness preserves convergence rates up to a quantifiable bias. We demonstrate the practical relevance of iPULA on a medical image reconstruction task, where proximal operators cannot be computed exactly. Experiments demonstrate the effectiveness of iPULA and support our theoretical results.