← Back to Sessions

Denoising Score Matching for Online Change Point Detection

Description: 

We study the problem of online change-point detection, where the goal is to identify abrupt changes in the underlying data distribution from a continuous stream of observations. The change-point can occur unexpectedly and is often challenging to detect when the distributions before and after the change are unknown. Unlike classical methods that typically rely on estimating probability densities for detection tasks, we propose an approach that focuses on estimating the score functions of the underlying distributions. By leveraging the score representation to construct detection statistics, we avoid estimating the exact form of the density function, which can be intractable in real-world scenarios. We study the theoretical performance of the proposed detection method, and validate its performance on both synthetic and real-world sequential seismic data streams.

Session: Building and Decoding High-resolution Earthquake Catalogs With Statistical and Machine-learning Tools - I

Type: Oral

Date: 4/15/2025

Presentation Time: 04:45 PM (local time)

Presenting Author: Liyan

Student Presenter: No

Invited Presentation: Yes

Poster Number:

Authors