文献类型：期刊文献

英文题名：Communication-efficient personalized federated learning with evolutionary strategies and cluster-aware knowledge transfer

作者：Wang, Yue[1,2];Hu, Han[1,2];Du, Wenli[1,2,3]

机构：[1]East China Univ Sci & Technol, State Key Lab Ind Control Technol, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[3]Huzhou Inst Ind Control Technol, Huzhou 313099, Peoples R China

年份：2025

卷号：328

外文期刊名：KNOWLEDGE-BASED SYSTEMS

收录：;EI(收录号：20253419008611);WOS:【SCI-EXPANDED(收录号:WOS:001584117500003)】；

基金：This work was supported by the National Key Research and Development Program of China (2022YFB3305900) , National Natural Science Foundation of China (Key Program: 62136003) , the State Key Laboratory of Industrial Control Technology, China (Grant No. ICT2024A24) and Fundamental Research Funds for the Central Universities.

语种：英文

外文关键词：Communication efficiency; Personalized federated learning; Evolutionary strategy; Knowledge transfer

摘要：Federated Learning (FL) enables distributed model training across multiple clients while preserving data privacy. However, in real-world scenarios, challenges such as high communication cost and non-iid data distributions often lead to degraded model performance and increased system overhead. While existing research has made some progress in addressing these challenges, achieving a balance between model accuracy and communication efficiency remains difficult. To address this issue, we propose FedECT, an efficient FL framework that leverages evolutionary perturbation strategies and cluster-aware knowledge transfer. Instead of transmitting high-dimensional gradients, FedECT employs fitness-based communication, significantly reducing bandwidth consumption while enhancing privacy protection. Specifically, clients generate perturbation model populations by evolutionary strategy and compute fitness values to transmit model information, which are then utilized for similarity-based clustering. Within each cluster, we introduce an optimization enhancement parameter that facilitates personalized knowledge transfer among similar clients, improving model generalization across heterogeneous data distributions. Extensive experiments on four datasets demonstrate that FedECT achieves up to 7% higher accuracy and reduces communication costs by at least 14% compared to most of the advanced FL approaches. Additionally, ablation studies confirm the effectiveness of fitness transmission based on evolutionary strategy and knowledge transfer within clusters in enhancing model performance and communication costs.