Ordering for Communication-Efficient Quickest Change Detection in a Decomposable Graphical Model

A quickest change detection problem is considered in a sensor network with observations whose statistical dependency structure across the sensors before and after described by decomposable graphical model (DGM). Distributed computation methods for this are proposed that capable of producing optimum centralized test statistic. The DGM leads to proper way collect nodes into local groups equivalent cliques graph, such clique statistic which summarizes all data can be computed within each clique. statistics transmitted decision maker produce In order further improve communication efficiency, an ordered transmission approach where transmissions fusion center then adaptively halted when sufficient information accumulated. This procedure always guaranteed provide optimal performance, despite not transmitting from cliques. lower bound on average number saved provided case seldom occurs approaches approximately half well behaved distance measure between distributions sufficiently large. We also extend graph different under hypothesis. Numerical results show significant savings using validate theoretical findings.