Abstract: While federated learning enables collaborative model training with preserved data locality, it remains vulnerable to evolving backdoor attacks that exploit its distributed architecture.