PFedMR: Adaptive Model Reorganization Framework

Federated Learning (FL), a distributed machine learning framework designed to preserve data privacy, is frequently constrained in inference performance and personalized adaptation due to the dual challenges of data heterogeneity and privacy preservation. To address these limitations, pfedmr is proposed, in which mutual learning and ensemble learning are adaptively integrated to mitigate the adverse effects caused by mismatched local data characteristics acrossclients. Additionally, experimental validation is conducted on multiple Non-IID datasets. It is demonstrated that the proposed method not only surpasses existing state-of-the-art approaches in terms of global model performance and inter-client fairness but also achieves a superior balance between personalized customization andmodel performance.