Multi-agent quality of experience control

Abstract

In the framework of the Future Internet, the aim of the Quality of Experience (QoE) Control functionalities is to track the personalized desired QoE level of the applications. The paper proposes to perform such a task by dynamically selecting the most appropriate Classes of Service (among the ones supported by the network), this selection being driven by a novel heuristic Multi-Agent Reinforcement Learning (MARL) algorithm. The paper shows that such an approach offers the opportunity to cope with some practical implementation problems: in particular, it allows to face the so-called “curse of dimensionality” of MARL algorithms, thus achieving satisfactory performance results even in the presence of several hundreds of Agents.