Sequential Change Point Detection via Denoising Score Matching

Zhou, Wenbin; Xie, Liyan; Peng, Zhigang; Zhu, Shixiang

Permalink

https://hdl.handle.net/2142/130270 Copy

Description

Title

Sequential Change Point Detection via Denoising Score Matching

Author(s)

• Zhou, Wenbin
• Xie, Liyan
• Peng, Zhigang
• Zhu, Shixiang

Issue Date

2025-09-17

Keyword(s)

• Quickest change detection
• Data-driven methods
• Denoising score matching

Abstract

Sequential change-point detection plays a critical role in numerous real-world applications, where timely identification of distributional shifts can greatly mitigate adverse outcomes. Classical methods often assume that the pre- and post-change distributions are either fully known or belong to a parametric family with unknown parameters, limiting their effectiveness for high-dimensional and complex data streams. In this paper, we propose a score-based CUSUM-type detection procedure applicable when the pre- and post-change distributions are unknown and not restricted to parametric families. We begin by estimating their score functions through noise injection and denoising score matching. The online detection statistic is then constructed by accumulating the differences between the pre- and post-change Hyvarinen score functions, and a change is declared once this statistic exceeds a pre-specified threshold. We consider both offline and online versions of score estimation. Through theoretical analysis, we demonstrate that denoising score matching can enhance detection power by effectively controlling the injected noise scale. Finally, we validate the practical efficacy of our method through numerical experiments on two synthetic datasets and a real-world earthquake precursor detection task, demonstrating its effectiveness in real applications.

Publisher

Allerton Conference on Communication, Control, and Computing

Series/Report Name or Number

2025 61st Allerton Conference on Communication, Control, and Computing Proceedings

ISSN

2836-4503

Type of Resource

Text

Genre of Resource

Conference Paper/Presentation

Language

eng

Handle URL

https://hdl.handle.net/2142/130270

Copyright and License Information

Copyright 2025 owned by the authors.

Owning Collections