A Hierarchical Bayes Approach for Distributed Binary Classification in Cyber-Physical and Social Networks

In this paper we consider a network of agents that can evaluate each other according to an interaction graph modeling some physical interconnection or social relationship. Each agent provides a score for its (out-)neighboring agents in the interaction graph. The goal is to design a distributed protocol, run by the agents themselves, to group the network nodes into two classes (binary classification) on the basis of the evaluation outcomes. We propose a hierarchical Bayesian framework in which the agents' belonging to one of the two classes is assumed to be a probabilistic event with unknown parameter. Exploiting such a hierarchical framework, we are able to design a distributed classification scheme in which nodes cooperatively classify their own state. We characterize the solution for a fault-diagnosis context in cyber-physical systems, and for an opinion-classification/community-discovery setup in social networks.