【摘要】 联邦学习作为新兴的分布式机器学习框架，能够为6G网络提供高效、安全的分布式数据资源使用方案。旨在研究最小化无线异构联邦学习系统的收敛挂钟时间。首先，给出了独立采样下联邦学习的收敛界限；进一步，提出带宽自适应分配方法以解决“掉队者”问题；最后，构建并求解最小化系统总挂钟时间的优化问题，得到最优采样概率。实验结果表明，该方法显著减少了联邦学习的收敛挂钟时间，验证了其实用性和优越性。更多还原

【Abstract】 Federated learning (FL),as an emerging distributed machine learning framework,provides an efficient and secure solution for utilizing distributed data resources in 6G networks.This paper aims to minimize the convergence clock time of heterogeneous wireless federated learning systems.Firstly,a convergence bound of FL is provided under independent sampling.Then,a bandwidth-adaptive allocation method is proposed to address the "straggler" issue.Finally,an optimization problem is formulated to minimize the total system clock time to obtain the optimal sampling probability.Experimental results demonstrate that the proposed method significantly reduces the convergence clock time for federated learning,validating its practicality and superiority.更多还原