Multiagent Learning

One of the greatest difficulties about multiagent learning is that the environment is not stationary with respect to the agent. In case of single agent learning problems, the agent has to maximize its expected reward with respect to an environment which is stationary. In case of multiagent scenarios, all the agents learning simultaneously poses a problem of non-stationarity in the environment which other agents have to take into account while computing their optimal behavior in such situations. One of the popular frameworks of addressing the problem of multiagent learning is the framework of stochastic games (SG) introduced by Shapeley [9]. In the following section we would emphasize on the SG framework and some of the well known algorithms that try to address the problem of multiagent learning in the SG Model.