PFedCS: A Personalized Federated Learning Method for Enhancing Collaboration among Similar Classifiers

• DOI: 10.1609/aaai.v39i20.35460
• Publication Date: 2025-04-11T13:09:08Z
• Authors: Siyuan Wu; Yongzhe Jia; Bowen Liu; Haolong Xiang; Xiaolong Xu; Wanchun Dou

Personalized federated learning (PFL) has recently gained significant attention for its capability to address the poor convergence performance on highly heterogeneous data and lack of personalized solutions traditional (FL). Existing mainstream approaches either perform aggregation based a specific model architecture leverage global knowledge or achieve personalization by exploiting client similarities. However, former overlooks discrepancies in distributions indiscriminately aggregating all clients, while latter lacks fine-grained collaboration classifiers relevant local tasks. In view this challenge, we propose Federated method Enhancing Collaboration among Similar Classifiers (PFedCS), which aims at improving client’s accuracy Concretely, it is achieved leveraging awareness classifier similarities above problems. By iteratively measuring distance parameters between clients clustering with each as cluster center, central server adaptively identifies collaborating similar distributions. addition, distance-constrained designed generate customized collaborative guide training. As result, extensive experimental evaluations conducted three datasets demonstrate that our achieves state-of-the-art performance.