FedBAP: Backdoor Defense via Benign Adversarial Perturbation in Federated Learning