Along with the unprecedented development of artificial intelligence (AI), a considerable number of intelligent applications are universally recognized to significantly facilitate the evolution of anthropogenic activities. The abundant AI computing power is one of the main pillars to fuel the booming of ubiquitous AI applications. As the computing power proliferates to a multitude of network edges, even end devices, the networking function bridges the gap, on the one hand, among ends-edges-clouds, on the other hand, between the multiple AI computing power and the heterogeneous AI requirements. The emerging new opportunities have spawned the deep integration between computing and networking. However, the complete development of the integrated system is under-addressed, including adaptability, flexibility, and profitability. In this article, we propose a computing-power networking framework for ubiquitous AI by establishing Networking in AI computing-power pool, denoted as Net-in-AI. We design the framework to enable the adaptability for computing-power users, the flexibility for networking, and the profitability for computing-power providers. We then formulate a computing-networking resource allocation problem, with the joint perspective of these three aspects. Experimental results prove the superior performance of the proposed framework in comparison to the current popular schemes.

中文翻译：

---

Net-in-AI：无处不在的AI具有适应性，灵活性和盈利能力的计算能力网络框架

随着人工智能（AI）的空前发展，公认的大量智能应用程序可以显着促进人为活动的发展。丰富的AI计算能力是推动无处不在的AI应用程序蓬勃发展的主要支柱之一。随着计算能力迅速扩展到甚至终端设备在内的众多网络边缘，联网功能一方面弥合了端到端云之间的差距，另一方面弥合了多个AI计算能力和异构AI之间的差距。要求。新兴的机遇催生了计算与网络之间的深度集成。但是，集成系统的完整开发未得到解决，包括适应性，灵活性和盈利能力。在本文中，我们通过在AI计算能力池中建立网络（称为Net-in-AI），为普适的AI提供一个计算能力网络框架。我们设计该框架以实现对计算能力用户的适应性，网络的灵活性以及对计算能力提供商的盈利能力。然后，从这三个方面的角度出发，我们提出了一个计算网络资源分配问题。实验结果证明，与当前流行的方案相比，该框架具有优越的性能。然后，从这三个方面的角度出发，我们提出了一个计算网络资源分配问题。实验结果证明，与当前流行的方案相比，该框架具有优越的性能。然后，从这三个方面的角度出发，我们提出了一个计算网络资源分配问题。实验结果证明，与当前流行的方案相比，该框架具有优越的性能。