Open access
2026
FedProRef: Communication-Efficient Prototype-Count Supervision for Federated Learning Under Severe Label Skew
Severe label skew leaves local classifier heads with insufficient supervision for underrepresented classes in federated learning (FL). We present FedProRef, an upload-efficient prototype-count supervision framework for frozen-feature, head-only FL. Each eligible client–class pair uploads only a prototype and sample cou...
Lei Wang, Hao Ren
· IEEE Access · 0 citations