Skip to content
Open access

Optimizing Value of Learning in Task-Oriented Federated Meta-Learning System Over NOMA Networks

2026 · IEEE Transactions on Wireless Communications · Vol 25, pp. 25795-25809 · 0 citations · 50 references

Abstract

A key limitation of federated learning (FL) is that its one-size-fits-all global model fails to provide customized solutions for the diverse tasks encountered in modern edge networks. Federated meta-learning (FML) offers a promising remedy by allowing each device to fine-tune the local model. However, the performance of FML can still be constrained by heterogeneous task requirements and resource costs, as well as limited communication efficiency. To address these challenges, we propose a task-oriented FML framework over non-orthogonal multiple access (NOMA) cluster-based networks. A novel metric, termed value of learning (VoL), is proposed to quantify the training utility of devices. Additionally, to prioritize the task of each device, a task-level weight (TLW) metric is defined. We aim to maximize the sum of TLW-based VoL across all devices. The formulated problem is non-convex. To solve it, we first analyze the two-device cluster case, where the problem is effectively solved using problem decomposition and convex optimization techniques. For the more complex case involving multiple devices per cluster, we develop a parameterized deep Q-network (PDQN)-based reinforcement learning algorithm to jointly optimize discrete and continuous variables. Simulation results for both scenarios demonstrate that the proposed approaches significantly outperform benchmark schemes, highlighting the effectiveness of the proposed solutions in enhancing FML performance.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.