A Privacy-Preserving Scheme for Federated Learning in a Cloud-Side-End Environment

Abstract

In recent years, protecting data privacy in distributed environments has become a key challenge with the rise of the Internet of Things (IoT) and edge computing. Federated learning helps mitigate the risk of data leakage by keeping data local, but issues such as authentication and insufficient privacy protection remain. In this paper, we propose a cloud-edge-end federated learning privacy protection scheme (FL-DPASS) that incorporates differential privacy and secret sharing. We introduce a Schnorr zero-knowledge proof authentication mechanism to secure end devices' access without revealing their identities. Additionally, we combine differential privacy and secret sharing to enhance security during model parameter transmission and aggregation. Experimental results demonstrate that our scheme strengthens data privacy protection while maintaining training efficiency and accuracy.