TS-FedNBS: Federated Edge Computing with Enhanced Robustness

Alina Basharat, Runhua Wang, Ping Xu

Abstract: Edge-assisted federated learning (FedEdge) that integrates an intermediate layer of edge nodes to reduce the workload for central server in traditional federated learning systems has been investigated in this work. However, the existing FedEdge mechanisms may be vulnerable to adversarial attackers. In this paper, we propose a two-stage robust aggregation scheme (TS-FedNBS) to enhance the resilience of FedEdge against Byzantine attackers. Specifically, TS-FedNBS employs a norm based screening (NBS) at the edge nodes and a median aggregation at the central server. Experimental results on real datasets indicate that the proposed method significantly enhances the resilience of FedEdge systems against Byzantine adversaries.