Nonparametric change detection and estimation in large-scale sensor networks

The problem of detecting changes in the distribution of alarmed sensors is considered. Under a nonparametric change detection framework, several detection and estimation algorithms are presented based on the Vapnik-Chervonenkis (VC) theory. Theoretical performance guarantees are obtained by providing error exponents for false-alarm and miss detection probabilities. Recursive algorithms for the efficient computation of test statistics are derived. The estimation problem is also considered in which, after detection is made, the location with maximum distribution change is estimated.