Though machine learning is widely used in wireless edge networks, the transmission of raw data still suffers from security and privacy leakage. Federated learning (FL) addresses these privacy concerns by enabling model training without sharing raw data. However, traditional centralized FL is vulnerable to a single point of failure. Blockchain-based federated learning (BFL) technology can provide FL with a more reliable and secure environment. In wireless edge networks with limited resources, BFL systems encounter challenges related to computing demands and network transmission overhead. To address these issues, we propose a BFL framework for wireless edge networks, which includes local client training, a consensus process, and edge server aggregation. A client selection policy is designed to exclude low-quality clients that could degrade training efficiency and accuracy. Additionally, a joint client selection and resource allocation scheme is implemented to optimize the allocation of computing and bandwidth resources necessary for BFL training and consensus. Simulation results demonstrate that the proposed approach improves BFL system accuracy while reducing delay.
Yang et al. (Mon,) studied this question.