B0462
Title: Sequential change detection via backward confidence sequences
Authors: Shubhanshu Shekhar - Carnegie Mellon University (United States) [presenting]
Aaditya Ramdas - Carnegie Mellon University (United States)
Abstract: A simple reduction from sequential estimation to sequential changepoint detection (SCD) is presented. In short, suppose the interest is in detecting changepoints in some parameter or functional $\theta$ of the underlying distribution. It is demonstrated that if a confidence sequence (CS) can be constructed for $\theta$, then SCD can be also successfully performed for $\theta$. This is accomplished by checking whether two CSs, one forward and the other backwards, ever fail to intersect. Since the literature on CSs has been rapidly evolving recently, the reduction provided immediately solves several old and new change detection problems. Further, the "backward CS", constructed by reversing time, is new and potentially of independent interest. Strong nonasymptotic guarantees are provided on the frequency of false alarms and detection delay, and demonstrate numerical effectiveness on several problems.