FedIoT: Optimizing the Communication-Efficient Federated Learning Aggregation Algorithm Under Heterogeneous Data for Large-Scale IoT

Training machine learning models effectively under severe communication, computation, and privacy restrictions has become more challenging due to the fast proliferation of Internet of Things (IoT) networks. With federated learning, dispersed IoT devices may train together without exposing any sensitive data. Unfortunately, privacy concerns, device heterogeneity, and communication constraints are common issues with large-scale installations. Through the utilization of adaptive asynchronous training, dynamic communication optimization, hierarchical aggregation, and differential privacy enhancement, we present a new federated learning communication framework that can withstand device delays, minimize bandwidth usage, improve scalability through edge-cloud coordination, and guarantee the protection of user data. Extensive testing on both real-world and synthetic datasets shows that the suggested architecture drastically cuts down on communication cost and training delay in heterogeneous IoT environments, all while keeping model accuracy high and privacy resilient.