FedMTFI: Feature Importance Based Optimized Multi Teacher Knowledge Distillation in Heterogeneous Federated Learning Environment
Researchers have introduced FedMTFI, a new architecture designed to improve federated learning in heterogeneous environments. This approach clusters clients based on similar hardware and model types, allowing each cluster to train a specialized model on non-IID data. The server then aggregates these models into prototypes that act as teachers for a global student model, enhanced by feature importance weighting using Shapley values for better accuracy and interpretability. AI
IMPACT Enhances federated learning for heterogeneous environments, potentially improving privacy-preserving AI development.