HEAFed:同态加密的高效通信异步联邦学习算法OA

Federated learning systems face risks of privacy data theft by attackers or semi-honest servers,and also suffer from low accuracy and inefficiency in unstable communication environments.To address these issues,this paper proposed an effi-cient communication asynchronous federated learning algorithm based on homomorphic encryption,named HEAFed.The algo-rithm firstly utilized the Chinese remainder theorem(CRT)to compress client parameter updates.It then employed an im-proved Paillier algorithm to encrypt the optimized parameter updates,ensuring user privacy security.Furthermore,it intro-duced an asynchronous aggregation mechanism with parameter adjustment,effectively incorporating the training outcomes of delayed clients.Experiments were conducted on the MNIST and CIFAR-10 datasets.Results show that HEAFed outperforms traditional privacy-preserving federated learning methods in both accuracy and efficiency.It performs especially well in unsta-ble communication environments.Compared with four baseline algorithms,HEAFed improves accuracy by 12.59％to 43.09％.