Compound Sequential Change-point Detection in Parallel Data Streams

We consider sequential change-point detection in parallel data streams, where each stream has its own change point. Once a is detected stream, this deactivated permanently. The goal to maximize the normal operation of pre-change while controlling proportion post-change streams among active at all time points. Taking Bayesian formulation, we develop compound decision framework for problem. A procedure proposed that uniformly optimal procedures which control expected also investigate asymptotic behavior method when number grows large. Numerical examples are provided illustrate use and performance method.