Minimum Error Resampling


Habibi, S, Drousiotis, E, Varsi, A ORCID: 0000-0003-2218-4720 and Maskell, S ORCID: 0000-0003-1917-2913 (2026) Minimum Error Resampling .

[thumbnail of Minimum_Error_Resampling (4).pdf] Text
Minimum_Error_Resampling (4).pdf - Author Accepted Manuscript
Available under License Creative Commons Attribution.
Download (519kB) | Preview

Abstract

Sequential Monte Carlo (SMC) samplers are Bayesian inference methods that employ sampling and resampling to generate a population of N weighted samples to estimate the state of a statistical model. This approach is robust and has made SMC commonly used across numerous areas of Statistics and Machine Learning. However, the resampling step alters the distribution of the sample weights, increasing the estimates’ uncertainty. This paper presents Minimum Error Resampling (MER), a resampling scheme which aims at minimizing the Mean Squared Error (MSE) in the distribution of the sample weights before and after resampling. Our theoretical analysis proves that MER achieves this minimum MSE. We present experimental results, in the context of sampling N Bayesian Decision Trees, showing that our proposed MER improves the performance of an SMC sampler when compared to the same sampler using other state-of-the-art resampling methods. The code is available on GitHub as open source.

Item Type: Conference Item (Unspecified)
Uncontrolled Keywords: 46 Information and Computing Sciences, 4611 Machine Learning, Networking and Information Technology R&D (NITRD), Generic health relevance
Divisions: Faculty of Health & Life Sciences
Faculty of Health & Life Sciences > Inst. Population Health
Depositing User: Symplectic Admin
Date Deposited: 26 Sep 2025 08:18
Last Modified: 17 Jun 2026 13:56
DOI: 10.1007/978-3-032-21477-5_16
Related Websites:
URI: https://livrepository.liverpool.ac.uk/id/eprint/3194606
Disclaimer: The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate.
Altmetric
CORE (COnnecting REpositories)