Unifying Sequential Monte Carlo with Resampling Matrices

TL;DR

This paper introduces a unified theoretical framework using resampling matrices to compare and analyze different resampling schemes in Sequential Monte Carlo algorithms, enabling better understanding and optimization of their performance.

Contribution

The paper develops a novel framework based on resampling matrices for comparing resampling schemes in SMC, providing new error formulas and insights.

Findings

Identifies the resampling scheme with minimal possible error.

Derives new asymptotic error formulas for resampling schemes.

Provides a unified approach to compare resampling methods in SMC.

Abstract

Sequential Monte Carlo (SMC) is a class of algorithms that approximate high-dimensional expectations of a Markov chain. SMC algorithms typically include a resampling step. There are many possible ways to resample, but the relative advantages of different resampling schemes remains poorly understood. Here, a theoretical framework for comparing resampling schemes is presented. The framework uses resampling matrices to provide a simple description for the SMC resampling step. The framework identifies the matrix resampling scheme that gives the lowest possible error. The framework leads to new asymptotic error formulas that can be used to compare different resampling schemes.