Diffusion differentiable resampling

TL;DR

This paper introduces a novel differentiable resampling method for sequential Monte Carlo that leverages an ensemble score diffusion model, providing theoretical consistency and improved empirical performance in filtering and high-dimensional image modeling.

Contribution

It presents a new diffusion-based resampling technique that is instantly differentiable and theoretically consistent, outperforming existing methods in various benchmarks.

Findings

Outperforms state-of-the-art differentiable resampling methods

Provides a theoretically consistent resampling distribution

Achieves competitive end-to-end performance in high-dimensional image modeling

Abstract

This paper is concerned with differentiable resampling in the context of sequential Monte Carlo (e.g., particle filtering). We propose a new informative resampling method that is instantly differentiable, based on an ensemble score diffusion model. We theoretically prove that our diffusion resampling method provides a consistent resampling distribution, and we show empirically that it outperforms the state-of-the-art differentiable resampling methods on multiple filtering and parameter estimation benchmarks. Finally, we show that it achieves competitive end-to-end performance when used in learning a complex dynamics-decoder model with high-dimensional image observations.