面向无线联邦学习网络的用户选择与资源分配联合优化  

Joint Optimization of Client Selection and Resource Allocation for Wireless Federated Learning Networks

作　　者：周钊辉 李奕彤 孙兆展 ZHOU Zhaohui;LI Yitong;SUN Zhaozhan(School of Electrical and Information Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学电气与信息工程学院,郑州450001

出　　处：《电讯技术》2024年第8期1275-1282,共8页Telecommunication Engineering

基　　金：国家自然科学基金青年基金项目(61801433)。

摘　　要：为了提高无线联邦学习网络性能,提出了一种用户选择和资源分配的联合优化算法。针对无线联邦学习网络中存在的资源受限和资源异质的问题,研究了长期能量约束下无线联邦学习网络的用户选择、通信资源分配和计算资源分配的联合优化问题,以达到最大化学习效率的目的。基于李雅普诺夫理论,将构建的长期优化问题转化为一系列的短期优化问题,并使用迭代算法优化用户选择和资源分配。与基准算法相比,所提算法能使学习效率提升10%以上,能在满足能量约束的条件下提高测试精度。In order to improve the performance of wireless federated learning networks,a joint optimization algorithm for client selection and resource allocation is proposed.Focusing on the resource-constrained and resource-heterogeneous problems in wireless federated learning networks,the joint optimization of client selection,communication resource allocation,and computing resource allocation for wireless federated learning networks is studied under the long-term energy constraints,in order to maximize learning efficiency.Based on Lyapunov theory,the long-term optimization problem is transformed into a series of short-term optimization problems,and an iterative algorithm is employed to optimize client selection and resource allocation.Compared with benchmark algorithms,the proposed algorithm can improve the learning efficiency by more than 10%and increase test accuracy while satisfying energy constraints.

关 键 词：无线网络 用户选择 联邦学习 资源分配 学习效率 

分 类 号：TN929[电子电信—通信与信息系统]