Decentralized sequential detection

A class of decentralized sequential detection problems is investigated. Under appropriate independence assumptions, it is shown that at each time instant the optimal local strategies are given by threshold tests on the likelihood of ratios. Furthermore, local decisions depend not only on the present and past observations, but on the past local decision as well. That is, for each local processor a different threshold exists for every different sequence of past decisions. Examples, computational techniques, discussions of the difficulties at hand, and suggestions for further exploration of the problem are presented.< >