An online learning approach to QoE-fair distributed rate allocation in multi-user video streaming

A Decentralized Partially Observable Markov Decision Process (DEC-POMDP) model of the resource allocation problem in multi-user video streaming over the Internet is formulated and developed to achieve both efficiency and fairness in terms of the expected value of end-user´s Quality of Experience (QoE) over a finite horizon. A distributed multi-agent reinforcement learning algorithm is proposed as a low-complexity solution to this decision problem. It is shown that the proposed learning algorithm would converge to the optimal value function and optimal policies of DEC-POMDP.