Intelligent Agent Querying

Given agents which have learned the transition functions for a discrete MDP, we examen how another agent, with no information about the state space, can efficiently determine how to query the other agents for knowledge of the MDP. This includes determining when to stop querying, as maximize the reward minus the penalty for each query (the reward in the MDP will increased due to more knowledge as a results of more queries, but so will the penalty). We also develop a method that determines what agents and what parts of the state space to query and find approaches with promising results for all problems, but most still need some adaptation before they are applicable to real situations.