A Kernel Multiple Change-point Algorithm via Model Selection

TL;DR

This paper introduces a kernel-based change-point detection method with a new penalty and theoretical guarantees, capable of identifying distributional changes in complex data.

Contribution

It proposes a novel penalty for kernel change-point detection, extending existing methods, and provides a non-asymptotic oracle inequality with new concentration results.

Findings

Accurately detects distributional change-points in synthetic data.

Effective even when mean and variance are unchanged.

Theoretical guarantees support practical performance.

Abstract

We tackle the change-point problem with data belonging to a general set. We build a penalty for choosing the number of change-points in the kernel-based method of Harchaoui and Capp{\'e} (2007). This penalty generalizes the one proposed by Lebarbier (2005) for one-dimensional signals. We prove a non-asymptotic oracle inequality for the proposed method, thanks to a new concentration result for some function of Hilbert-space valued random variables. Experiments on synthetic data illustrate the accuracy of our method, showing that it can detect changes in the whole distribution of data, even when the mean and variance are constant.