摘要
分布式网络人工智能模型协同技术是分布式数据安全流通的底层支撑,是实现6G网络智慧内生的关键技术.对6G网络中的模型协同技术进行了全面的综述.回顾了中心侧大模型中,数据并行、模型并行及混合并行等模型协同技术.面对边缘智能中边缘节点算力及内存不足、通信带宽受限的问题,从拆分学习、联邦学习、拆分联邦学习、分布式群体学习等几个方面探讨总结了边缘智能的模型协同技术.最后,强调了在6G网络中,为了不重复训练相似的大模型,需要大小模型协同进化.
Abstract
Distributed artificial intelligence(AI)model collaboration techniques serve as the foundational support for secure distributed data circulation and are critical for enabling intelligent capabilities in 6G networks.This paper provides a comprehensive survey of model collaboration techniques within 6G networks.It reviews collaborative techniques in centralized large-scale models,including data parallelism,model parallelism,and hybrid parallelism.To address the challenges in edge intelligence,such as limited computational power,memory constraints,and restricted communication bandwidth at edge nodes,this study explores and summarizes model collaboration approaches such as split learning,federated learning,split federated learning,and distributed collective learning.Finally,the paper highlights the necessity of collaborative evolution between large and small models in 6G networks to avoid redundant training of similar large-scale models.
维普
万方数据