Anomaly Search Over Many Sequences With Switching Costs

This paper considers the quickest search problem to identify anomalies among large numbers of data streams. These streams can model, for example, disjoint regions monitored by a mobile robot. A particular challenge is version in which experimenter must suffer cost each time stream being sampled changes, such as robot spend moving between regions. In this paper, we propose an algorithm accounts switching costs varying confidence threshold that governs when switches new stream. Our main contributions are easily computable approximations both optimal value and parameter determines flagged anomaly, using Brownian motion approximations. Further, empirically show (i) uniform improvement interest (ii) roughly equivalent performance small comparing closest available algorithm.