Lancelot: Towards Efficient and Privacy-Preserving Byzantine-Robust Federated Learning within Fully Homomorphic Encryption

• DOI: 10.48550/arxiv.2408.06197
• Publication Date: 2024-08-12
• Authors: Siyang Jiang; Hao Yang; Qipeng Xie; Chuan Ma; Sen Wang; Guoliang Xing
• Keywords: Homomorphic Encryption

In sectors such as finance and healthcare, where data governance is subject to rigorous regulatory requirements, the exchange utilization of are particularly challenging. Federated Learning (FL) has risen a pioneering distributed machine learning paradigm that enables collaborative model training across multiple institutions while maintaining decentralization. Despite its advantages, FL vulnerable adversarial threats, poisoning attacks during aggregation, process typically managed by central server. However, in these systems, neural network models still possess capacity inadvertently memorize potentially expose individual instances. This presents significant privacy risk, attackers could reconstruct private leveraging information contained itself. Existing solutions fall short providing viable, privacy-preserving BRFL system both completely secure against leakage computationally efficient. To address concerns, we propose Lancelot, an innovative efficient framework employs fully homomorphic encryption (FHE) safeguard malicious client activities preserving privacy. Our extensive testing, which includes medical imaging diagnostics widely-used public image datasets, demonstrates Lancelot significantly outperforms existing methods, offering more than twenty-fold increase processing speed, all