A hierarchical federated-learning method that lets the trained model grow larger and stronger as knowledge moves from resource-constrained end devices up through edge servers to the cloud, instead of forcing every tier to share the weakest device's model size. Students learn how bridge-sample online distillation transfers knowledge between differently-sized models on neighboring tiers, and how a self-rectification step keeps the growing cloud model from converging to a suboptimal solution under heterogeneous, non-IID data and node mobility.
FedEEC is a hierarchical federated-learning framework for the end-edge-cloud setting, where AI models are trained collaboratively across resource-constrained end devices, more capable edge servers, a…