Sequential change detection via backward confidence sequences

TL;DR

This paper introduces a novel method for sequential change detection by leveraging confidence sequences, including a new backward confidence sequence, providing strong guarantees and demonstrating effectiveness across various problems.

Contribution

It proposes a reduction from sequential estimation to change detection using confidence sequences, introducing the concept of backward CSs for improved detection capabilities.

Findings

Strong nonasymptotic guarantees on false alarms and detection delay

Effective numerical performance on multiple change detection problems

Introduction of a new backward confidence sequence

Abstract

We present a simple reduction from sequential estimation to sequential changepoint detection (SCD). In short, suppose we are interested in detecting changepoints in some parameter or functional $\theta$ of the underlying distribution. We demonstrate that if we can construct a confidence sequence (CS) for $\theta$, then we can also successfully perform SCD for $\theta$. This is accomplished by checking if two CSs -- one forwards and the other backwards -- ever fail to intersect. Since the literature on CSs has been rapidly evolving recently, the reduction provided in this paper immediately solves several old and new change detection problems. Further, our "backward CS", constructed by reversing time, is new and potentially of independent interest. We provide strong nonasymptotic guarantees on the frequency of false alarms and detection delay, and demonstrate numerical effectiveness on several problems.