信息通信技术与政策

信息通信技术与政策

信息通信技术与政策 ›› 2021, Vol. 47 ›› Issue (4): 93-96.doi: 10.12267/j.issn.2096-5931.2021.04.017

所属专题: 5G

上一篇    

基于5G 边缘计算的深度学习架构与应用

A deep learning framework based on 5G edge computing with its applications

李晓民   

  1. 中国移动(成都)产业研究院,成都 610064
  • 出版日期:2021-04-15 发布日期:2021-04-28
  • 作者简介:
    李晓民:中国移动(成都)产业研究院技术规划部副总经理,主要从事5G 教医农商等垂直行业产品及技术应用研究工作

LI Xiaomin   

  1. China Mobile Industry Research Institute, Chengdu 610064, China
  • Online:2021-04-15 Published:2021-04-28

摘要: 针对深度神经模型在网络边缘难以训练的问题,构建了一种基于5G 边缘计算的深度学习模型训练架构。架构利用5G 边缘计算接入网打通边缘智能设备与边缘计算层的数据通信,模型训练过程采用各边缘计算节点利用本地数据进行全模型训练,再由中心服务器进行模型参数汇集和更新的分布式训练模式,既保证了模型训练的数据集多样性,又减少了网络压力和保障了本地数据隐私,是一种非常具有潜力的深度学习边缘计算架构。

关键词: 5G, 边缘计算, 深度学习, 边缘智能, 医疗边缘云

Abstract: Aiming to mitigate the training difficulty of deep neural models at the edge of network, this paper introduces a deep learning model training framework based on 5G edge computing. The presented training framework connects smart devices and performs data communication via the 5G edge computing access network. The model training is firstly implemented in edge computing nodes using local data, and then the model parameters are collected and globally updated in the central server. This scenario ensures the diversity of data sets for training, reduces network􀆳s pressure and protects privacy of local data. Thus, it is a promising edge computing architecture for deep learning.

Key words: 5G, edge computing, deep learning, edge intelligence, medical edge cloud