Safeguarding cross-silo federated learning with local differential privacy

• DOI: 10.1016/j.dcan.2021.11.006
• Publication Date: 2021-11-30T23:28:05Z
• Authors: Chen Wang; Xinkui Wu; Gaoyang Liu; Tianping Deng; Kai Peng; Shaohua Wan
• Keywords: Differential Privacy; Federated Learning; Safeguarding; Privacy Protection

Federated Learning (FL) is a new computing paradigm in privacy-preserving Machine (ML), where the ML model trained decentralized manner by clients, preventing server from directly accessing privacy-sensitive data clients. Unfortunately, recent advances have shown potential risks for user-level privacy breaches under cross-silo FL framework. In this paper, we propose addressing issue using three-plane framework to secure FL, taking advantage of Local Differential Privacy (LDP) mechanism. The key insight here that LDP can provide strong protection while still retaining user statistics preserve its high utility. Experimental results on three real-world datasets demonstrate effectiveness our