A Lightweight Decentralized Federated Learning Framework for the Industrial Internet of Things

Abstract

Federated learning (FL) has recently gained significant attention in edge computing, the Industrial Internet of Things (IIoT), and Internet of Things (IoT) due to its ability to enable distributed clients to train models collaboratively while keeping the original data local. However, existing works usually suffer from limited communication resources, dynamic network conditions, and heterogeneous client properties, which hinder effective FL in IIoT scenarios. To address the above challenges simultaneously, we propose a Lightweight Decentralized Federated Learning Framework for the Industrial Internet of Things (LDFLF). LDFLF uses the ternary quantization technique to compress the client model, reduce the communication overhead, and improve model transmission efficiency. Experiments show the proposed method’s superiority in communication efficiency, model accuracy, and convergence speed, making it particularly suitable for resource-constrained IIoT environments. Compared to traditional federated learning methods, LDFLF framework achieves an average communication cost reduction of 80% and an average model accuracy improvement of 5.3% on IID data and 10.2% on Non-IID data, while significantly accelerating the convergence speed.