Efficiency-enhanced Blockchain-based Client Selection in Heterogeneous Federated Learning

Abstract

In Federated Learning (FL), blockchain has been extensively used to achieve distributed and tamper-resistant data processing. However, typical Blockchain-based Federated Learning (BFL) rarely considers clients’ resource and computing limits. High-capacity clients may be sacrificed when all clients train on the same neural network. This paper proposes a Blockchain-based Heterogeneous Federated Learning (BlocFL) model to address the challenges above. BlocFL replaces the central server with a consortium blockchain, and several neural networks are employed for local training. Considering the challenges in resource allocation in BFL, especially in heterogeneous networks, we propose a consortium blockchain-based heterogeneous federated learning client selection method. The proposed method optimizes the choice of client nodes under the limits of computational resources. Experiment results demonstrate that our method can allocate appropriate neural network models to each client and effectively improve the efficiency of local training in HFL. It also can achieve a comparable level of accuracy to the baseline approach with similar training parameters.