FedSPA Uses Shapley Values to Shield Federated Learning From Poisoning Attacks
FedSPA combines Shapley-value analysis and model parameter similarity to identify malicious clients in federated learning under data poisoning attacks.
FedSPA combines Shapley-value analysis and model parameter similarity to identify malicious clients in federated learning under data poisoning attacks.
A six-client CIFAR-10 experiment found that layered federated-learning defenses improved accuracy and resilience to poisoning and adversarial attacks, while privacy ...
Simulated distribution-grid case studies found that a serverless, committee-based federated-learning framework combined near-centralized false-data-injection detection with resistance to tested malicious ...
A systematic review of 128 studies finds that secure federated learning in medical IoT requires layered defenses that balance poisoning ...
Researchers at Donghua University have developed RFLDefense, a layer-wise defense framework that suppresses backdoor attacks in federated learning even when ...
Researchers have developed GRAB-FL, a graph-aware federated learning framework that uses graph neural networks to assign trust scores to client ...
A new study introduces auditable update-utility certificates that verifiably measure the value of each client's contribution in personalized federated learning ...
© 2026 Scienmag - Science Magazine
© 2026 Scienmag - Science Magazine