Normalizing self-supervised learning for provably reliable Change Point Detection

TL;DR

This paper introduces a spectral normalization-based representation learning approach for change point detection, providing theoretical guarantees and outperforming existing methods on standard datasets.

Contribution

It integrates spectral normalization with deep representation learning for CPD and offers the first provable reliability framework for such methods.

Findings

Significantly outperforms state-of-the-art CPD methods on three datasets.

Provides theoretical guarantees for the embeddings used in CPD.

Demonstrates the effectiveness of spectral normalization in enhancing CPD reliability.

Abstract

Change point detection (CPD) methods aim to identify abrupt shifts in the distribution of input data streams. Accurate estimators for this task are crucial across various real-world scenarios. Yet, traditional unsupervised CPD techniques face significant limitations, often relying on strong assumptions or suffering from low expressive power due to inherent model simplicity. In contrast, representation learning methods overcome these drawbacks by offering flexibility and the ability to capture the full complexity of the data without imposing restrictive assumptions. However, these approaches are still emerging in the CPD field and lack robust theoretical foundations to ensure their reliability. Our work addresses this gap by integrating the expressive power of representation learning with the groundedness of traditional CPD techniques. We adopt spectral normalization (SN) for deep representation learning in CPD tasks and prove that the embeddings after SN are highly informative for CPD. Our method significantly outperforms current state-of-the-art methods during the comprehensive evaluation via three standard CPD datasets.