Data-Efficient Quickest Outlying Sequence Detection in Sensor Networks

Abstract

A sensor network is considered where at each sensor a sequence of random variables is observed. At each time step, a processed version of the observations is transmitted from the sensors to a common node called the fusion center. At some unknown point in time the distribution of observations at an unknown subset of the sensor nodes changes. The objective is to detect the outlying sequences as quickly as possible, subject to constraints on the false alarm rate, the cost of observations taken at each sensor, and the cost of communication between the sensors and the fusion center. Minimax formulations are proposed for the above problem and algorithms are proposed that are shown to be asymptotically optimal for the proposed formulations, as the false alarm rate goes to zero. It is also shown, via numerical studies, that the proposed algorithms perform significantly better than those based on fractional sampling, in which the classical algorithms from the literature are used and the constraint on the cost of observations is met by using the outcome of a sequence of biased coin tosses, independent of the observation process.