A Personalized Federated Learning Algorithm Based on Dynamic Weight Allocation

• DOI: 10.3390/electronics14030484
• Publication Date: 2025-01-27T14:42:23Z
• Authors: Yazhi Liu; Siwei Li; Wei Li; Hui Qian; Haonan Xia

Federated learning is a privacy-preserving distributed machine paradigm. However, due to client data heterogeneity, the global model trained by traditional federated averaging algorithm often exhibits poor generalization ability. To mitigate impact of some existing research has proposed clustered learning, where clients with similar distributions are grouped together reduce interference from dissimilar clients. since distribution unknown, determining optimal number clusters difficult, leading reduced convergence efficiency. address this issue, paper proposes personalized based on dynamic weight allocation. First, each allowed obtain tailored fit its local distribution. During aggregation process, server first computes similarity updates between and dynamically allocates weights models these similarities. Secondly, use received exclusive train their via algorithm. Extensive experimental results demonstrate that, compared other algorithms, method effectively improves accuracy speed.