Privacy-preserving approach to edge federated learning based on blockchain and fully homomorphic encryption

To address the issues concerned with high risk of single-point failure, weak privacy protection and poor resistance to poisoning attacks in edge federated learning, An edge federatied learning privacy protection scheme based on blockchain and full homomorphic encryption is proposed. This approach utilises blockchain to provide edge federatied learning with the characteristics of anti-tampering ,anti-single-point failure and data process transparency combining the CKKS full homomorphic encryption scheme to encrypt relevant computational parameters, thus reduce the risk of privacy leakage. Additionally, an unsupervised model parameter update identification mechanism is designed, using the consistency between historical model updates of edge servers as the identification basis, and enhances the accuracy of aggregation model while identifying malicious edge server updates. The experimental results demonstrate that the proposed method is capable of resisting 70% of poisoning attacks from malicious edge servers while providing privacy protection, encryption, and maintaining the integrity of the blockchain simultaneously. Furthermore, it is able to achieve high model accuracy and meets the stringent requirements for security, accuracy, and traceability commonly associated with edge federated learning scenarios.