Byzantine-Resilient Distributed Hypothesis Testing With Time-Varying Network Topology

In this article, we study the problem of distributed hypothesis testing over a network mobile agents with limited communication and sensing ranges to infer true collaboratively. particular, consider scenario where there is an unknown subset compromised that may deliberately share altered information undermine team objective. We propose two algorithms each agent maintains updates sets beliefs (i.e., probability distributions hypotheses), namely local xmlns:xlink="http://www.w3.org/1999/xlink">actual (LB AB, respectively, for brevity). both algorithms, at every time step, shares its AB other within range makes local observation update LB. Then, can use shared ABs under certain conditions. One requires receiving number xmlns:xlink="http://www.w3.org/1999/xlink">each instant ; accumulates xmlns:xlink="http://www.w3.org/1999/xlink">over time after exceeds prescribed threshold. Otherwise, rely on agent’s current LB new AB. prove mild assumptions noncompromised converges almost surely hypothesis, without requiring connectivity in underlying time-varying topology. Using simulation unmanned aerial vehicles aiming classify adversarial among themselves, illustrate compare proposed algorithms. Finally, show experimentally second algorithm consistently outperforms first terms speed convergence.