基于联邦学习的半异步分层设备协调调度方案

A semi‐asynchronous hierarchical device coordination scheduling scheme based on federated learning

为了应对边缘计算存在的数据隐私性和高通信延迟问题,提出一种基于联邦学习的半异步分层设备协调调度方案。通过引入联邦学习机制,实现了边缘计算应用下的数据隐私保护。此外,针对传统联邦学习因通信资源短缺导致的用户交互时延过大的问题,设计了一种云、边和终端设备分层通信架构,通过采用云-边半异步通信和边-终端同步通信架构来设计一个联合边缘节点关联和资源分配问题,并提出了一种交替方向乘子法和块坐标更新(alternating direction method of multipliers and block coordinate update,ADMM-BCU)融合算法来找到次优解,最终实现训练准确度和传输延迟之间的均衡。仿真结果证明所提出的协调调度方案能够有效提升系统性能。

To deal with the data privacy and high communication latency problems in edge computing,a semi�asynchronous hierarchical device coordination scheduling scheme based on federated learning(FL)is proposed.By introducing the FL mechanism,data privacy protection is achieved in edge computing applications.In addition,a cloud,edge and terminal device hierarchical communication architecture is designed to address the problem of excessive user interaction latency caused by the shortage of communication resources in traditional FL.A joint edge node association and resource allocation problem is designed by adopting a cloud-edge semi�asynchronous communication and an edge-terminal synchronous communication architecture,and an alternating direction method of multipliers and block coordinate update(ADMM-BCU)algorithm is proposed to find the sub-optimal solution and finally achieve a balance between training accuracy and transmission delay.Simulation results prove that the proposed coordinated scheduling scheme can effectively improve the system performance.