Social learning and controlled sensing

Multiagent social learning deals with the problem of Bayesian estimation of an underlying state when agents can use private observations together with local decisions of previous agents to infer the state. How can controlled sensing be performed at a global level when local agents perform social learning? This paper considers two such examples motivated by statistical signal processing applications in sequential detection. The examples show that social learning can yield unusual behavior - in stopping problems, the stopping set is non-convex and also the optional policy can have a multi-threshold structure.