Distance-Aware Hierarchical Federated Learning in Blockchain-Enabled Edge Computing Network

Federated learning (FL) has been proposed as an emerging paradigm to perform privacy-preserving distributed machine learning in the Internet of Things (IoT). However, the communication overhead caused by partial model aggregations will increase the model training latency. In this article, a multilayer blockchain-enabled hierarchical FL (HFL) network is proposed for low-latency model training while ensuring data security. Meanwhile, we theoretically analyze the bottleneck of the model accuracy with the total data distance due to the imbalanced data distribution. Moreover, the mathematical expression of the model error with respect to IoT devices (IDs) association and local data distribution is provided, then the upper bound of the model error is represented by the total data distance. To further improve the learning performance, the distance-aware HFL (DAHFL) algorithm is investigated, which optimizes ID association strategy based on dual-distance, and allocates computing and communication resources alternatively. Finally, the working process of the blockchain-enabled HFL system is exhibited by the blockchain simulation platform and the efficiency of the proposed DAHFL algorithm is demonstrated by the simulation results.