无线电通信技术2024，Vol.50Issue(4)：696-703,8.DOI:10.3969/j.issn.1003-3114.2024.04.011
基于自适应聚合时间的半同步联邦资源优化算法
Semi-synchronous Federated Resource Optimization Algorithm Based on Adaptive Aggregation Time
摘要
Abstract
Federated learning is an efficient distributed machine learning method,in which multiple devices use their own local data for distributed model training.There is no need to exchange local data.Instead,it only need to build a shared global model by exchan-ging model parameters,thereby protecting the user's privacy.In order to balance model performance and communication delay,in the semi-synchronous federated learning scenario,an optimization problem that minimizes the weighted sum of model performance and ag-gregation time is established using weight parameters.Optimization variables include the aggregation time for global model updates,user scheduling,and the bandwidth and transmit power of participating upload users.The proposed Mixed Integer Non-Linear Programming(M1NLP)problem is decomposed into two sub-problems to solve using Alternating Direction Method of Multipliers(ADMM).Simula-tion experiments prove that the proposed algorithm can reduce the aggregation time by 73％at the expense of 4％model performance,and significantly improve communication efficiency.关键词
物联网/半同步联邦学习/用户调度/资源优化Key words
internet of things/semi-synchronous federated learning/user scheduling/resource allocation分类
电子信息工程引用本文复制引用
李铁,庄琲,林尚静,韩志博..基于自适应聚合时间的半同步联邦资源优化算法[J].无线电通信技术,2024,50(4):696-703,8.基金项目
湖北省工程研究中心2024年开放课题The Opening Project of Hubei Engineering Research Center in 2024 （）
