A game theoretic framework for communication decisions in multiagent systems

Communication is the process of transferring information between multiple entities. We study the communication process between different entities that are modeled as multiple agents. To study the interactions between these entities we use Game theory, which is a mathematical tool that is used to model the interactions and decision process of these multiple players/agents. In the presence of multiple players their interactions are generally modelled as a stochastic game. Here we assume that the communication medium, protocols and language are already present in this multiagent system. We address the question of information selection in the communication process. In this thesis, we develop a formal framework for communication between different agents using game theory. Our major contributions are: A classifications of the multi agent systems and what information to communicate in these various cases. Algorithms for Inverse Reinforcement Learning in multiagent systems, which allow an agent to get a better understanding about the other agents. A mathematical framework using which the agents can make two important decisions, when to communicate, and, more importantly what to communicate in different classes of multiagent systems.