Generative diffusion posterior sampling for informative likelihoods

TL;DR

This paper introduces a new diffusion posterior SMC sampler that improves sampling efficiency for generative diffusion models, especially with informative likelihoods or outliers, by leveraging correlated observation paths.

Contribution

It proposes a novel diffusion posterior SMC method that enhances statistical efficiency through correlated observation paths, addressing limitations of previous approaches.

Findings

Improved sampling efficiency under outlier conditions.

Enhanced performance with highly informative likelihoods.

Empirical validation confirms efficiency gains.

Abstract

Sequential Monte Carlo (SMC) methods have recently shown successful results for conditional sampling of generative diffusion models. In this paper we propose a new diffusion posterior SMC sampler achieving improved statistical efficiencies, particularly under outlier conditions or highly informative likelihoods. The key idea is to construct an observation path that correlates with the diffusion model and to design the sampler to leverage this correlation for more efficient sampling. Empirical results conclude the efficiency.