Jiafeng Zhao

Privacy-Preserving Rényi Layer-Wise Budget Allocation Against Gradient Leakage for Federated Learning

2025 · IEEE Transactions on Mobile Computing

Federated learning (FL) is vulnerable to gradient-based privacy attacks, where malicious attackers reconstruct training data from exchanged gradients. While existing differential privacy (DP) defenses mitigate this, they often cause excessive additive noise due to the …