A Multi-Agent Reinforcement Learning Based Approach to Quality of Experience Control in Future Internet Networks

In the perspective of the emerging Future Internet framework, the Quality of Experience (QoE) Control functionalities are aimed at approaching the desired QoE level of the applications by dynamically selecting the most appropriate Classes of Service supported by the network. In the present work, this selection is driven by Multi-Agent Reinforcement Learning, namely by the Friend-Q learning algorithm. The proposed dynamic approach differs from the traffic classification approaches found in the literature, where a static assignment of Classes of Service to application instances is performed. All these improvements are aimed at adding a cognition loop to telecommunication networks, by making use of Multi-Agent Reinforcement Learning, and at fostering the intelligent connectivity between applications and networks.