Title: Two-phase privacy-preserving scheme for federated learning in edge networks

Authors: Hongle Guo; Yingchi Mao; Xiaoming He; Jie Wu

Addresses: College of Computer and Information, Hohai University, Nanjing, Jiangsu, 211100, China; Key Laboratory of Water Big Data Technology of Ministry of Water Resources, Hohai University, Nanjing, Jiangsu, 211100, China ' College of Computer and Information, Hohai University, Nanjing, Jiangsu, 211100, China; Key Laboratory of Water Big Data Technology of Ministry of Water Resources, Hohai University, Nanjing, Jiangsu, 211100, China ' College of Computer and Information, Hohai University, Nanjing, Jiangsu, 211100, China; Key Laboratory of Water Big Data Technology of Ministry of Water Resources, Hohai University, Nanjing, Jiangsu, 211100, China ' Center for Networked Computing, Temple University, Philadelphia, PA 19122, USA

Abstract: In this article, to protect sharing parameters from being leaked and reduce time cost or low federated learning accuracy, we propose a two-phase privacy-preserving scheme for federated learning (TPPP-FL). Our method works in two phases. Specifically, in the uploading phase, we present a many-to-one HE algorithm to protect model parameters. In the downloading phase, to reduce the size of the keys, the zero-knowledge signature scheme is improved, and then a one-to-many zero-knowledge digital signature scheme is proposed to ensure the integrity and irreversibility of model parameters. Two datasets (MNIST and ORL) and two training models (CNN and MLP) have been set in experiments. Theoretical analysis and experimental results show that the TPPP-FL can effectively reduce time costs without losing the accuracy of the models compared to the same type of other federated learning schemes.

Keywords: federated learning; privacy-preserving; one-to-many; many-to-one; encryption homomorphic; zero knowledge proof.

DOI: 10.1504/IJSNET.2023.132540

International Journal of Sensor Networks, 2023 Vol.42 No.3, pp.170 - 182

Received: 12 Apr 2023
Accepted: 13 Apr 2023
Published online: 27 Jul 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article