面向多任务联邦学习的移动设备调度方法  

Mobile Devices Scheduling Method for Multi-task Federated Learning

作　　者：焦翔 魏祥麟 范建华 薛羽[1] 贾茹娜 JIAO Xiang;WEI Xianglin;FAN Jianhua;XUE Yu;JIA Runa(School of Computer and Software,Nanjing University of Information Science and Technology,Nanjing 210044,China;The 63rd Research Institute,National University of Defense Technology,Nanjing 210007,China)

机构地区：[1]南京信息工程大学计算机与软件学院,南京210044 [2]国防科技大学第六十三研究所,南京210007

出　　处：《指挥与控制学报》2024年第1期88-99,共12页Journal of Command and Control

摘　　要：边缘指挥控制场景下,将多任务联邦学习中的移动设备调度问题建模为多目标优化问题,在模型收敛性、传输可靠性、计算资源有限性约束下,最小化联邦学习任务每轮的训练时延和能耗。为了求解该0-1整数规划问题,提出了一种基于差分进化的移动设备调度算法,将调度方案作为个体,通过交叉变异迭代进化得到具有最佳适应度的次优解。仿真结果表明,所提算法能够在保证模型准确率的前提下,有效降低训练过程中的时延与能量消耗。In edge command and control scenario,the mobile device scheduling problem in multi-task federated learning(FL)is modelled as a multi-objective optimization problem.Under the constraints of model convergence,transmission reliability,and limited computation resources,the optimization goal is to minimize the training delay and energy consumption of each round of multiple FL tasks.To solve the 0-1 integer programming problem,a mobile device scheduling algorithm based on differential evolution is proposed.The proposed algorithm treats each feasible scheduling policy as an individual,and obtains a sub-optimal solution with optimum fitness through crossover and mutation iterative evolution.Simulation results show that the proposed algorithm can ensure the accuracy of the model,and can effectively reduce the time and energy consumption in the training process.

关 键 词：多任务联邦学习 移动边缘计算 时延 能量消耗 

分 类 号：E9[军事]