Efficient Federated Meta-Learning Over Multi-Access Wireless Networks

Federated meta-learning (FML) has emerged as a promising paradigm to cope with the data limitation and heterogeneity challenges in today’s edge learning arena. However, its performance is often limited by slow convergence corresponding low communication efficiency. In addition, since available radio spectrum IoT devices’ energy capacity are usually insufficient, it crucial control resource allocation consumption when deploying FML practical wireless networks. To overcome challenges, this paper, we rigorously analyze contribution of each device global loss reduction round develop an algorithm (called NUFM) non-uniform selection scheme accelerate convergence. After that, formulate problem integrating NUFM multi-access systems jointly improve rate minimize wall-clock time along cost. By deconstructing original step step, devise joint strategy solve theoretical guarantees. Further, show that computational complexity can be reduced from $O(d^{2})$ notation="LaTeX">$O(d)$ (with model dimension notation="LaTeX">$d$ ) via combining two first-order approximation techniques. Extensive simulation results demonstrate effectiveness superiority proposed methods comparison existing baselines.