Split Federated Learning for Resource-Constrained Edge Computing Networks

doi:10.1109/TCE.2025.3626460

Split Federated Learning for Resource-Constrained Edge Computing Networks

2025-10-28

0

PRE

AI

DOI：10.1109/TCE.2025.3626460

Original

Original request for help

Save

Abstract

En 中文

Compute First Networking (CFN) integrates distributed computing resources across cloud, edge, and device layers to meet diverse computing demands. Within this context, Split Federated Learning (SFL) has emerged as an effective approach to reduce device workload and enhance privacy by integrating model splitting into federated training. However, deploying SFL in edge computing networks is challenging due to device heterogeneity, unreliable channels, and the trade-off between convergence and resource usage. This paper formulates a joint optimization problem that considers split layer selection, computing resource allocation, and transmission power control to minimize the total training latency required to reach a target model accuracy. We propose an efficient Block Coordinate Descent (BCD)-based algorithm to solve this problem. Experimental results demonstrate that the proposed framework significantly reduces the total training latency and adapts effectively to varying wireless conditions.

Keywords:

Edge computing networks

split federated learning

model splitting

transmission power control

computing resources allocation

Split Federated Learning for Resource-Constrained Edge Computing Networks

Abstract

Journal

Organization