Tag
This paper presents a reconstructed benchmark analyzing federated aggregation methods under model poisoning and backdoor attacks, finding Trimmed Mean most accurate in clean settings and Krum most robust under attacks, while auditing metric implementations and noting reproducibility caveats.