Decentralized Federated Learning in IoT Environments: A Hierarchical Approach

Decentralized federated learning has emerged as a promising approach to train machine learning models in Internet of Things (IoT) environments, where data privacy, communication constraints, and resource limitations are critical concerns. This article presents a hierarchical architecture for decentralized federated learning in IoT, designed to address the unique challenges of IoT devices. The proposed architecture comprises multiple levels, including the edge level, local aggregators, regional aggregators, and a global aggregator. Each level plays a distinct role in facilitating communication and coordination while considering resource constraints. Furthermore, we provide an in-depth analysis of the convergence properties of the proposed framework, leveraging gradient and derivative relations. Experimental results demonstrate the effectiveness and efficiency of the proposed approach in achieving accurate and privacy-preserving learning in IoT settings.