Tag
This paper introduces BackDFL, a unified benchmark for systematically evaluating backdoor attacks and defenses in decentralized federated learning, revealing critical failure modes in current robustness methods.