Hierarchical Federated Learning Privacy Protection Framework with Enhanced Privacy and Resistance to Byzantine Attacks
Guilin Guan, Ting Zhi, Huimin Cai, Yang Cao, Hongtao Xie · 2024
As a new computing paradigm, edge computing provides computing services at the edge of the network. Compared with the traditional cloud computing model, it has the characteristics of high reliability and low latency. Federated learning, as a distributed machine learning method, although it has the characteristics of protecting privacy and data security, still faces problems such as device heterogeneity and data imbalance, resulting in long training time and low training efficiency at some edge ends. To address the aforementioned shortcomings, we propose a federated learning incentive mechanism scheme that enhances privacy in edge scenes and is resistant to Byzantine attacks. Our scheme can encrypt weights through multi-key homomorphic encryption methods to resist data recovery attacks initiated by malicious edge servers and malicious devices. Simultaneously, based on marginal loss technology, detect malicious clients or upload low-quality clients, and evaluate the contribution level of edge devices using Shapley technology as the basis for distributing rewards to participants. To address unauthorized user access, a hybrid password system for identity authentication has been designed, ensuring that only authenticated participants can participate in the training process of federated learning. In addition, extensive experiments were conducted on three datasets to evaluate the performance of our scheme. The results show that our scheme outperforms other advanced federated learning methods in terms of accuracy, fairness, and robustness.