<ul><li>Prototype-based federated learning has emerged as a promising approach for knowledge transfer among clients with data heterogeneity.</li><li>In this paper, a framework named Federated Learning via Semantic Anchors (FedSA) is proposed to address the issues caused by statistical and model heterogeneity.</li><li>FedSA introduces semantic anchors as prototypes and uses anchor-based regularization and classifier calibration to ensure consistent representations and decision boundaries across clients.</li><li>Experiments demonstrate that FedSA outperforms existing prototype-based federated learning methods on various classification tasks.</li></ul>

FedSA: A Unified Representation Learning via Semantic Anchors for Prototype-based Federated Learning

Discover more