Learning via active hypothesis testing over networks

摘要: This paper considers a problem of distributed active hypothesis testing. At every time instant, individual nodes in the network adaptively choose sensing action and receive noisy local (private) observations as outcomes. The distribution is parameterized by discrete parameter (hypotheses). marginals joint observation conditioned on each are known locally at nodes, but true parameter/hypothesis not known. An update rule analyzed which first possibly randomized function their past actions. Nodes then perform Bayesian belief (distribution estimate) based current observations. Each node communicates these updates to its neighbors, performs “non-Bayesian” linear consensus using log-beliefs neighbors. Under mild assumptions for general class selection strategies, we show that any wrong converges zero exponentially fast, exponential rate learning characterized nodes' influence average distinguishability between observations' distributions (randomized) under hypothesis.