<ul><li>Federated learning in Cyber-Physical-Social Systems (CPSS) integrates differential privacy to tackle privacy risks.</li><li>Existing research focuses on adjusting noise levels or discarding gradients to address differential privacy noise but neglects noise removal hindering convergence.</li><li>A novel framework is proposed for differentially private federated learning that corrects noisy gradients using a server-side mechanism, ensuring privacy guarantees without compromising accuracy.</li><li>Evaluation on benchmark datasets shows that this framework outperforms others in maintaining privacy while achieving top performance.</li></ul>

GCFL: A Gradient Correction-based Federated Learning Framework for Privacy-preserving CPSS

Discover more