Monte Carlo off-policy reinforcement learning: a rough set approach

This paper introduces a rough set approach to reinforcement learning by cooperating agents using a variation of the Monte Carlo off-policy control method. The problem considered in this article is how to measure the value of a state relative to the collections of similar behaviors and select an optimal policy. The solution to this problem is made possible by considering behavior patterns of swarms in the context of approximation spaces, which provide a framework for computing rough inclusion values for weights in estimating the value of a swarm state. Two different forms of the Monte Carlo off-policy reinforcement learning method are considered as a part of a study of learning in real-time by a swarm. The contribution of this article is the presentation of a new Monte Carlo off-policy control method defined in the context of approximation spaces. The ecosystem provided by swarms of zebra danio fish (Brachydanio Rerio) has been selected to facilitate study of reinforcement learning. This ecosystem is briefly described. In addition, the results of experiments using reinforcement learning techniques to simulate swarm behavior of the zebra danio ecosystem for two forms of the Monte Carlo off-policy control method are given.