Adaptive Federated Learning with High-Efficiency Communication Compression
Xuyang Xing, Honglei Liu · 2024
This paper thoroughly investigates two major challenges faced by Federated Learning (FL): high communication costs and adaptive requirements during the learning process. Based on this issue, by combining hybrid compression technology with FedYogi adaptive learning strategies, we provide an innovative algorithm named Adaptive Federated Learning with HighEfficiency Communication Compression (AFL-HECC), which significantly reduces communication overhead while achieving adaptability in the learning process. In a non-convex stochastic optimization setting, we demonstrate that AFL-HECC can achieve a convergence rate of $\mathcal{O}\left(\frac{1}{\sqrt{T K m}}\right)$. Finally, simulation experiments are given to validate the effectiveness of the proposed algorithm.