FedDPKD: Federated learning with dual-phase knowledge distillation for label distribution skew

Puntos clave

• The model shows improved accuracy over traditional methods, particularly in datasets with label distribution skew.
• Key improvements include a 15% increase in model performance on skewed datasets, enhancing its practical application.
• Federated learning techniques were employed with dual-phase knowledge distillation to address label distribution issues.
• This strategy may enable better cooperation among decentralized data sources, though further validation is needed.