Data aggregation method for privacy protection in federated learning environment
Jianfei Xiao, Shu Wang · 2023
With the development and application of computer federated learning technology, there are more attack methods targeting federated learning, posing challenges to data privacy security. This article proposes a weight hiding federated learning secure aggregation method to address the risk of inference attacks caused by leaked aggregation weights during the federated learning aggregation process. This method prevents attackers or malicious servers from obtaining information such as aggregation weights, and utilizes Bayesian differential privacy to reduce noise disturbance to the model. Experiments have shown that this method can effectively prevent attackers from implementing inference attacks through malicious tampering with aggregated weights, and significantly improve the accuracy and efficiency of federated learning algorithms.