Resilient distributed hypothesis testing under time-varying multi-agent networks with multiple types of adversarial agents

In this paper, the resilient distributed hypothesis testing problem under multi-agent networks is studied, where the normal agents that are not attacked aim to cooperatively learn a true state from a set of hypotheses. Different from the existing works, the considered network is heterogeneous and time -varying, and it is allowed that multiple types of agents can be attacked. Within this framework, by designing a novel filtering mechanism, an effective resilient distributed algorithm is developed to solve the considered hypothesis testing problem. In the convergence analysis of the algorithm, a key definition about the robustness of heterogeneous time-varying networks is firstly introduced, then by combining the designed filtering mechanism with a delay-based analysis approach, it is proven that the beliefs of the true state held by all normal agents converge to 1 under the proposed algorithm, which means that the normal agents successfully learn the true state. Finally, the theoretical findings are verified through simulations, and a comparison is provided to show that the proposed algorithm by utilizing the network heterogeneity may tolerate more adversarial agents than the algorithm designed in homogeneous networks.(c) 2023 Elsevier B.V. All rights reserved.

Automated summary

In this paper, the resilient distributed hypothesis testing problem under multi-agent networks is studied. A novel filtering mechanism is designed and an effective resilient distributed algorithm is developed to solve the problem. The convergence analysis of the algorithm proves that the normal agents successfully learn the true state. Simulation results verify the theoretical findings and demonstrate the superiority of the proposed algorithm in tolerating adversarial agents.