A1715
Title: Multiple-experiment quickest change detection under cost constraints
Authors: Patrick Vincent Lubenia - University of Pittsburgh (United States) [presenting]
Taposh Banerjee - University of Pittsburgh (United States)
Abstract: Classic models of quickest change detection typically involve a single experiment used to monitor a stochastic process. This study considers the scenario where, at each observation time, an observer must select from a variety of experiments, each characterized by distinct information qualities and associated costs. The main objective is to minimize the worst-case average detection delay while satisfying constraints on false alarm rate and cost. To achieve this, the 2E-CUSUM algorithm is introduced for scenarios involving two experiments. The study then explores more complex designs involving multiple experiments, extending the 2E-CUSUM framework accordingly. The proposed algorithms are demonstrated to be asymptotically optimal.
