Accelerating the Computation of Solutions in Resource Allocation Problems Using an Evolutionary Approach and Multiagent Reinforcement Learning.

In systems composed by a high number of highly coupled components, aligning the optimum of the system with the optimum of those individual components can be conflicting, especially in situations in which resources are scarce. In order to deal with this, many authors have proposed forms of biasing the optimization process. However, mostly, this works for cooperative scenarios. When resources are scarce, the components compete for them, thus those solutions are not necessarily appropriate. In this paper a new approach is proposed, in which there is a synergy between: (i) a global optimization process in which the system authority employs metaheuristics, and (ii) reinforcement learning processes that run at each component or agent. Both the agents and the system authority exchange solutions that are incorporated by the other party. The contributions are twofold: we propose a general scheme for such synergy and show its benefits in scenarios related to congestion games.