Title: Distribution-Free Changepoint Localization Following Sequential Change Detection

Abstract:

This study presents a novel, distribution-free methodology for generating post-detection confidence intervals for changepoints, specifically following the termination of a sequential change detection process. While it is established that conformal test martingales are effective for sequentially identifying distributional shifts, they inherently lack the capability to infer the specific timing of the detected change. Previous approaches to post-detection inference were constrained by the necessity of specifying pre- and post-change distribution classes; in contrast, this work achieves changepoint localization without imposing any distributional assumptions. The paper derives finite-sample coverage guarantees, conditional upon the accuracy of the detection. Furthermore, it offers non-asymptotic bounds regarding the conditional expected size of these confidence sets. Under appropriate asymptotic conditions, the authors demonstrate that the conditional expected size of the confidence set remains uniformly bounded. Empirical evaluations on both simulated and real-world datasets confirm the framework's robust performance. To the best of our knowledge, this represents the first general distribution-free framework for sequential changepoint localization that ensures valid post-detection coverage guarantees.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC