ABSTRACT
Sequential hypothesis testing in multiple data streams (e.g., sensors, populations, multichannel systems) has important practical applications. Some of the applications include public health, genomics, environmental monitoring, military defense, and cyber security. This chapter considers the sequential hypothesis testing problem where observations are acquired one at a time, in a number of data streams, and where the number and location of “patterns” of interest are either completely or partially unknown a priori. It focuses on two multistream sequential tests, the Generalized Sequential Likelihood Ratio Test and the Mixture Sequential Likelihood Ratio Test, which are based on the maximum and average likelihood ratio over all possible hypotheses regarding the number and location of signals, respectively. The chapter establishes asymptotic optimality properties of the two proposed sequential tests for any possible signal configuration. It shows that in the case of uncoupled and independent streams both the tests are scalable with respect to the number of streams.
