Type-Sensitive Social Learning
Joni Shaska, Urbashi Mitra · 2023
The problem of distributed hypothesis testing with correlated observations is studied. Specifically, systems in which the behavior is governed by both the underlying hypothesis, as well as an underlying empirical distribution on the network state is considered. Thus, there is significant coupling between the interim decisions of the agents and the signals they transmit. The current model addresses increased coupling relative to prior work. The optimal decay rate for optimal detection is computed; key properties associated with this error rate are derived. The utility of the analysis is shown via the consideration of a multi-class problem wherein agents within each class have specific properties and interact with agents of other classes via signal enhancement or jamming. This multi-class case is studied numerically and it is shown that there is a optimal ratio between class populations that maximizes the decay rate of the error.