Search

Future Internet is expected to meet explosive traffic growth and extremely complex architecture, which tend to make the traditional NTC strategies inefficient and even ineffective. Inspired by the latest breakthroughs of AI and its power to address large-scale and complex difficulties, the network community has begun to consider shifting the NTC paradigm from legacy rule-based to novel AI-based. As an applied inter-discipline, design and implementation are important. Although there have been some preliminary explorations along this frontier, they are either limited by only envisioning the prospects, or too scattered to provide high-level insight into a general methodology. To this end, we start with the domain knowledge relationships of AI and NTC, summarizing a baseline workflow toward deep reinforcement learning, which will be the dominant method for the AI-NTC paradigm. On top of that, we argue that AI-NTC training and running must be carried out in online environments in closed-loop fashion for the purpose of putting ti into practice. A series of challenges and opportunities are discussed from a realistic viewpoint, and a set of new architecture and mechanism to enable the online and closed-loop AI-NTC paradigm are proposed. Hopefully, this work can help the AI community to better understand NTC and the NTC community to better live with AI. [ABSTRACT FROM AUTHOR]