移动通信2025，Vol.49Issue(3)：37-45,9.DOI:10.3969/j.issn.1006-1010.20250112-0001
无线算力网络中面向AI任务的分布式通信计算协同研究
Distributed Communication and Computing Collaboration for AI Tasks in Wireless Computing Power Networks
摘要
Abstract
By converging communication,computing,and model capabilities,6G networks are able to provide distributed,efficient,real-time,and energy-saving AI service capabilities.As its foundational platform,wireless computing power networks,through on-demand computing services and rational distributed communication-computing collaborative design,are poised to become an important support solution for the extremely high computing power demands of future mobile AI applications.This paper first classifies and analyzes typical collaborative modes,including centralized mode,hierarchical clustering mode,decentralized mode,and edge distributed collaborative mode.Second,it reviews the main distributed training and inference methods,covering federated learning,split learning,model optimization and collaborative inference,as well as multi-agent deep reinforcement learning.Finally,an evaluation metrics system is designed from aspects of AI task accuracy,latency,density,and energy efficiency,providing important references for performance research and assessment of mobile AI tasks.关键词
6G/移动AI任务/分布式训练/协同推理Key words
6G/mobile AI tasks/distributed training/collaborative inference分类
电子信息工程引用本文复制引用
张申虎,杨蕊侨,谢云菁,闫实..无线算力网络中面向AI任务的分布式通信计算协同研究[J].移动通信,2025,49(3):37-45,9.基金项目
国家自然科学基金"面向差异化服务的通信-感知-计算融合无线组网理论与方法"(62371067) （62371067）
