Performance and robustness in sequential analysis of stochastic data flows

Abstract

The problem of sequential testing of hypotheses on parameters of an observed stochastic data flow is considered. Simple and composite hypotheses setting are investigated for 3 families of data flows: homogeneous independent observations, independent observations with inhomogeneities, dependent observations forming Markov chains. The interest is focused on two issues: performance characteristics (error probabilities and mathematical expectation of the random number of observations) calculation, and robustness analysis of sequential tests under distortions (deviations from the hypothetical model assumptions). Approaches to solve these two problems are proposed and developed via construction of asymptotic expansions (w.r.t. the discretization parameter and the distortion levels) of the performance characteristics. The robust sequential procedures are constructed within families of generalized sequential test