通信网络与大模型的融合与协同

摘要：随着以大模型（LM）为代表的生成式人工智能（AI）的兴起，将大模型应用于通信网络的研究引起了学术界和工业界的广泛关注。回顾了目前大模型的主流神经网络架构及其能力涌现机理，然后从AI与通信的双向协同、网络大模型部署两方面，深入探讨了通信网络大模型研究的主要进展。还分析了网络大模型NetGPT将要面临的挑战以及未来的发展方向。考虑到基于AI/机器学习（ML）的通信模型相较于传统模型获得的出色性能，认为将通信网络与大模型进行融合并使二者协同工作，能进一步地提升系统的性能。要实现通信网络与大模型的融合与协同，本质上是要构建好网络大模型，云边协同就提供了一种很好的网络大模型部署方案。

关键词：LM；生成式AI；网络智能；NetGPT；模型协同

Abstract: Along with the springing up of generative artificial intelligence (AI), notably epitomized by large models (LM), the incorporation of these LMs within communication networks has attracted extensive attention in both academia and industry. An overview of the dominant deep neural network (DNN) architecture of LMs and its emerging capabilities is introduced. The significant advancements potential achieved by applying LMs for communication networks from two aspects are discussed, namely, the mutual collaboration between AI and communications, and the deployment of network generative pre-trained transformer (NetGPT). Additionally, the imminent challenges and further work are also discussed. Considering the outstanding performance of AI/machine learning (ML)-based communication models compared to traditional models, it is believed that integrating communication networks with large models and enabling them to work together can further enhance system performance. To realize the integration and collaboration of communication networks and large models, it is essentially necessary to build NetGPT properly. Edge-cloud collaboration provides a good deployment solution for NetGPT.

Keywords: LM; generative AI; network intelligence; NetGPT; model collaboration