CoASFL: Coupled Consistency Optimization for Asynchronous Split Federated Learning

Computer Science > Machine Learning arXiv:2610.26992 (cs) [Submitted on 19 Oct 2026] Title: CoASFL: Coupled Consistency Optimization for Asynchronous Split Federated Learning Authors: Junhao Cheng, Yunming Liao, Hongli Xu, Yang Xu, Zuan Xie.

计算机科学 > 机器学习 arXiv:2610.26992 (cs) [提交于 2026 年 10 月 19 日] 标题：CoASFL：异步分割联邦学习的耦合一致性优化 作者：Junhao Cheng, Yunming Liao, Hongli Xu, Yang Xu, Zuan Xie。

Abstract: Split Federated Learning (SFL) has emerged as a promising paradigm for training large-scale deep models on resource-constrained edge devices. Although asynchronous SFL alleviates the straggler bottleneck caused by system heterogeneity, it inevitably introduces two coupled dimensions of inconsistency: horizontal inconsistency across clients and vertical inconsistency between clients and the server.

摘要：分割联邦学习 (SFL) 已成为在资源受限的边缘设备上训练大规模深度模型的一种有前景的范式。尽管异步 SFL 缓解了由系统异构性导致的落后者瓶颈，但它不可避免地引入了两个耦合的一致性维度：跨客户端的水平不一致性以及客户端与服务器之间的垂直不一致性。

These inconsistencies, further exacerbated by statistical heterogeneity, result in biased activation representations, model divergence, and severe client drift. To address these challenges, we propose CoASFL, an inconsistency-aware asynchronous SFL framework that incorporates a dynamic activation merging mechanism and a consistency-aware model aggregation mechanism.

这些不一致性在统计异构性的进一步加剧下，导致了偏差的激活表示、模型发散以及严重的客户端漂移。为了应对这些挑战，我们提出了 CoASFL，这是一个具有一致性感知能力的异步 SFL 框架，它结合了动态激活合并机制和一致性感知模型聚合机制。

Specifically, CoASFL constructs more consistent activation sequences from heterogeneous clients to alleviate horizontal inconsistency, and dynamically adjusts aggregation weights according to client participation frequencies and progress gaps to mitigate vertical inconsistency. By jointly coordinating activation-level and model-level consistency, CoASFL effectively suppresses inconsistency propagation during asynchronous training.

具体而言，CoASFL 通过构建来自异构客户端的更一致的激活序列来缓解水平不一致性，并根据客户端参与频率和进度差距动态调整聚合权重，以减轻垂直不一致性。通过联合协调激活层级和模型层级的一致性，CoASFL 有效地抑制了异步训练过程中的不一致性传播。

Extensive experiments on an 80-device heterogeneous edge testbed demonstrate that CoASFL consistently outperforms state-of-the-art SFL approaches, accelerating training by 1.61$\times$ to 3.74$\times$ and improving final model accuracy by up to 3.26% under heterogeneous scenarios.

在 80 台设备的异构边缘测试平台上的大量实验表明，CoASFL 始终优于最先进的 SFL 方法，在异构场景下将训练速度提升了 1.61 倍至 3.74 倍，并将最终模型准确率提高了多达 3.26%。