Cluster-Based Client Selection for Dependent Multi-Task Federated Learning in Edge Computing

TL;DR

This work tackles the challenge of dependent multi-task federated learning (FL) in mobile edge computing (MEC) by presenting CoDa-FL, a framework that combines $EMD$-based client clustering with DAG-aware task scheduling via reinforcement learning. The authors derive a convergence bound showing the required rounds $R_v(\varepsilon_v)$ grow with intra-cluster heterogeneity $\Delta_{\mathcal{C}_i}$ and related variance terms, justifying cluster-based selection and dependency-aware allocation. They implement two key algorithms: an $EMD$-based clustering procedure and a PPO-based DAG scheduler, and validate performance gains on a four-task DAG with heterogeneous devices, achieving about a $5.05\%$ reduction in total completion time compared with baselines. The results demonstrate faster convergence, lower communication/computation costs, and higher accuracy, highlighting a scalable solution for efficient FL in edge environments with dependent tasks.

Abstract

We study the client selection problem in Federated Learning (FL) within mobile edge computing (MEC) environments, particularly under the dependent multi-task settings, to reduce the total time required to complete various learning tasks. We propose CoDa-FL, a Cluster-oriented and Dependency-aware framework designed to reduce the total required time via cluster-based client selection and dependent task assignment. Our approach considers Earth Mover's Distance (EMD) for client clustering based on their local data distributions to lower computational cost and improve communication efficiency. We derive a direct and explicit relationship between intra-cluster EMD and the number of training rounds required for convergence, thereby simplifying the otherwise complex process of obtaining the optimal solution. Additionally, we incorporate a directed acyclic graph-based task scheduling mechanism to effectively manage task dependencies. Through numerical experiments, we validate that our proposed CoDa-FL outperforms existing benchmarks by achieving faster convergence, lower communication and computational costs, and higher learning accuracy under heterogeneous MEC settings.

Figures (3)

• Figure 1: Cluster-based FL client selection in MEC.
• Figure 2: Comparison of training time by different methods.
• Figure 3: Accuracy vs. total time across four tasks.