Adding reasoning and cognition to the internet
Manolis Sifalakis, Manolis Mavrikis, George Maistros · Lancaster EPrints (Lancaster University) · 2004
Present computer networks are laid over a set of dummy boxes performing specialised tasks, following an outdated model that provides a best-effort type of service. Current application needs have pushed research towards the so-called networking, which aims to enable programmability/adaptability of the network fabric according to varying user requirements. This paper takes one step further and discusses a new topic for network research, which promotes the interplay between active network technology and artificial intelligence. It introduces a context of reasoning and cognition in the network fabric, based on reinforcement learning and other AI strategies that will allow training of the infrastructure with past experiences/history of events and user preferences. We analyse the requirements of this class of networks and identify basic building blocks of a potential architecture. The outcome of this research envisions dependable, resilient, self-manageable, selfconfigurable and survivable networks or what we call herein cognitive networks.