A two-stage federated learning method for personalization via selective collaboration

Open Webpage

Jiuyun Xu, Liang Zhou, Yingzhi Zhao, Xiaowen Li, Kongshang Zhu, Xiangrui Xu, Qiang Duan, Ruru Zhang

Published: 01 Jan 2025, Last Modified: 03 Oct 2025Comput. Commun. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•A two-stage framework to mitigate negative knowledge transfer in the FL system.•Innovatively break the limitation of the client’s fixed belonging category.•A new dynamic approach to select collaborative partners for clients.•A hierarchical model fusion method to promote knowledge transfer within groups.•Avoid frequent high-dimensional operations.