Distributed reinforcement learning for multi-agent sequential decision problems
Learning in multi-agent system is a challenging task because many of the characteristics of the learning and decision making problem in a single-agent setting are changed due to the interactions of the multiple agents. The problem becomes even more harder if the agents have to select a sequence of actions. The goal of this project is to design algorithms by which hierarchical learning automata can find Nash paths or optimal pats in sequential decision problems. If the number of stages becomes large, the agents must use a new strategy that bootstraps on obtained information.