A Consensus-based Approach for Distributed Quickest Detection of Significant Events in Networks

This paper studies the problem of quickest detection of significant events in networks, where nodes undergo a change in the data generating the distributions of their observations due to events that occurred at some unknown time. Events can propagate dynamically along edges in the network to affect more nodes over time; however, the propagation dynamics are assumed to be unknown. A consensus-based distributed-detection algorithm is proposed to detect a "significant" event, i.e., at least eta nodes have been affected by the event, as quickly as possible, subject to false alarm constraints. It is shown that the proposed distributed algorithm achieves an equivalent performance to that of a centralized algorithm, which was shown to be firstorder asymptotically optimal, as the false alarm rate goes to zero. Finally, numerical experiments are provided to evaluate the efficiency of the proposed algorithm.