CODEA : Online Container Deployment and Edge Association for Multi-Edge Federated Synergy Learning

Fu, Shucun and Dong, Fang and Xu, Xiaolong and Shen, Dian and Bilal, Muhammad and Dustdar, Schahram (2026) CODEA : Online Container Deployment and Edge Association for Multi-Edge Federated Synergy Learning. IEEE Transactions on Mobile Computing. pp. 1-18. ISSN 1536-1233

Text (CODEA_Online_Container_Deployment_and_Edge_Association_for_Multi-Edge_Federated_Synergy_Learning)
CODEA_Online_Container_Deployment_and_Edge_Association_for_Multi-Edge_Federated_Synergy_Learning.pdf - Accepted Version
Available under License Creative Commons Attribution.
Download (5MB)

Abstract

Federated edge learning (FEEL) is a prospective paradigm enabling edge devices to collaboratively participate in machine learning model training, unlocking countless opportunities for edge intelligence. As an extension of FEEL, federated synergy learning (FSyL) alleviates the computation and communication burdens on resource-constrained end-devices by offloading partial model layers to edge servers for synergistic training. However, existing work largely ignores the impact of budgeted multi-edge service deployment and dynamic device-server association on training performance, causing significant accuracy degradation and increased costs. To address these limitations, this paper investigates critical performance bottlenecks of executing FSyL and formulates a novel benefit maximization problem that jointly optimizes online container deployment and edge association. To efficiently tackle this intractable problem, we propose CODEA, an online COntainer Deployment and Edge Association framework based on the contextual multi-armed bandit model. CODEA guides deploying containerized FSyL services across multiple edge servers under a budget constraint, maximizing device coverage and training robustness before each global update stage. Following container deployment, the online edge association determines the device-server association during each local training round, maximizing cost-savings and ensuring the success rate of update collection. Extensive experiments demonstrate that CODEA significantly improves training accuracy and cost-savings compared to state-of-the-art methods.