Communication-efficient hierarchical federated learning for IoT heterogeneous systems with imbalanced data

Federated Learning (FL) is a distributed learning methodology that allows multiple nodes to cooperatively train deep model, without the need share their local data. It promising solution for telemonitoring systems demand intensive data collection, detection, classification, and prediction of future events, from different locations while maintaining strict privacy constraint. Due concerns critical communication bottlenecks, it can become impractical send FL updated models centralized server. Thus, this paper studies potential hierarchical in Internet Things (IoT) heterogeneous systems. In particular, we propose an optimized user assignment resource allocation over architecture IoT This work focuses on generic class machine are trained using gradient-descent-based schemes considering practical constraints non-uniformly across users. We evaluate proposed system two real-world datasets, show outperforms state-of-the-art solutions. Specifically, our numerical results highlight effectiveness approach its ability provide 4–6% increase classification accuracy, with respect consider distance-based assignment. Furthermore, could significantly accelerate training reduce overhead by providing 75–85% reduction rounds between edge server, same model accuracy.