Anomaly Search With Multiple Plays Under Delay and Switching Costs

The problem of searching for $L$ anomalous processes among notation="LaTeX">$M$ is considered. At each time, the decision maker can observe a subset notation="LaTeX">$K$ (i.e., multiple plays). measurement drawn when observing process follows one two different distributions, depending on whether normal or abnormal. goal to design policy that minimizes Bayes risk which balances between sample complexity, detection errors, and switching cost associated with across processes. We develop policy, dubbed consecutive controlled sensing (CCS), achieve this goal. On hand, by contrast existing studies sensing, CCS senses consecutively reduce cost. other controls operation in closed-loop manner switch necessary guarantee reliable inference. prove theoretically asymptotically optimal terms minimizing as error approaches zero complexity increases). Simulation results demonstrate strong performance finite regime well. Index Terms - Anomaly detection, active hypothesis testing, sequential experiments.