A Novel Decentralized Federated Split Learning Framework in 6G Edge Networks

Federated learning (FL) and split learning (SL) are two popular distributed machine learning approaches. As a paradigm of FL, decentralized federated learning (DFL) architecture enables edge devices to collaboratively train a shared machine learning model without a central parameter server. Meanwile, SL can reduce computing load of edge devices by model splitting and assignment. In this paper, we propose a distributed federated split learning (DFSL) framework. In terms of framework design, DFSL combines the advantages of SL and DFL, enabling resource-constrained edge devices to participate in the training of complex models while reducing the risks of network bottlenecks and single points of failure during training. In terms of model training, DFSL enhances the model testing accuracy compared to DFL and significantly reduces the training time compared to SL. We conduct detailed performance analysis of the proposed DFSL algorithm, demonstrating its efficiency and convergence in training. Numerical results validate the superiority of DFSL over DFL and SL in model training.