Privacy-Preserving Verifiable Asynchronous Federated Learning
Yuanyuan Gao, Lulu Wang, Lei Zhang · 2021
Federated learning (FL) is a recently proposed technique to cope with growing data and break the barriers among datasets, which enables nodes to train machine learning models without sharing their local datasets. However, the data privacy and model performance concerns in asynchronous federated learning hinder its deployment in practical applications, especially in dynamic scenarios. To address these problems, we propose a verifiable asynchronous federated learning with a peer-to-peer network based on local dataset test and cosine value examination to improve the model performance. We also design a privacy-preserving scheme by using the local differential privacy (LDP) to protect data privacy. We evaluate our scheme on the model accuracy and convergence performance. Numerical results show the high accuracy and efficiency of our proposed scheme while protecting privacy.