移动通信2024，Vol.48Issue(7)：66-72,7.DOI:10.3969/j.issn.1006-1010.20240628-0002
面向6G的时间敏感异构联邦边缘智能
Latency-Sensitive Heterogeneous Federated Edge Intelligence for 6G
摘要
Abstract
The deep integration of artificial intelligence and the realization of intelligent endogenous capabilities are critical evolutionary trends for 6G networks.Federated Edge Intelligence(FEI)is an emerging distributed collaborative training architecture that provides intelligent services to users while effectively protecting local data privacy.However,significant differences in data characteristics and available communication and computing resources among different edge nodes in computing networks result in high heterogeneity.Traditional federated learning architectures exhibit poor global model convergence rates and high service response latency under these conditions.This paper proposes a latency-sensitive heterogeneous federated edge intelligence system architecture that employs asynchronous model collaboration to mitigate the latency impact of lagging nodes.Additionally,based on data contribution analysis,a strategy combining low-performance edge node data discarding with load forwarding is introduced to further reduce the staleness effect on service response latency in asynchronous collaboration mechanisms.Experimental tests on a heterogeneous FEI prototype system based on Raspberry Pi demonstrate that the proposed scheme significantly reduces service response latency.关键词
智能内生/时间敏感网络/联邦学习/异构网络Key words
intelligent endogenous/latency-sensitive networks/federated learning/heterogeneous networks分类
信息技术与安全科学引用本文复制引用
王连,何升涛,蔡浩晖,李莹玉,肖泳,石光明..面向6G的时间敏感异构联邦边缘智能[J].移动通信,2024,48(7):66-72,7.基金项目
国家自然科学基金"面向6G群体智能资源共享博弈基础理论研究"(62071193),"面向智能语义理解的计算成像方法研究"(61976169),"语义通信基础理论与方法研究"(62293483),"基于多通道压缩感知的高分辨高动态范围红外成像方法研究"(61871304) （62071193）
鹏城实验室重大攻关项目(PCL2021A12) （PCL2021A12）
中央高校基本科研业务费资助,HUST"基于联邦学习的数字孪生网络建模研究"(2023JYCXJJ029) （2023JYCXJJ029）
中国地质大学(武汉)"地大学者"人才岗位科研启动经费资助(2021164) （武汉）
湖北省国际科技合作计划项目"面向印尼通信工程复杂场景的云边协同智能运维系统研发及应用"(2023EHA009) （2023EHA009）
