Sharif Digital Repository

One of the challenging problems in artificial intelligence is credit assignment which simply means distributing the credit among a group, such as a group of agents. We made an attempt to meet this problem with the aid of the reinforcement learning paradigm. In this paper, expertness framework is defined and applied to the multi-agent credit assignment problem. In the expertness framework, the critic agent, who is responsible for distributing credit among agents, is equipped with learning capability, and the proposed credit assignment solution is based on the critic to learn to assign a proportion of the credit to each agent, and the used proportion should be learned by reinforcement... 

Multi-agent systems (MAS) is a solution to the nowadays encountered problems, which have the characteristics such as distributiveness, dynamism and the need to adaptation, robustness, efficiency, and reusability. This paper proposed a solution to multi-agent credit assignment problem. The contribution is to equip the critic agent (who is responsible for distributing reinforcements among agents) with learning capability. Some criteria are used to propose an inner feedback to the critic. Results of simulation show the applicability of the method to a task, which has the characteristic that the agent has to decide from a large set of actions. The research is a preliminary step to more in-depth... 

In this paper, a novel reinforcement learning method inspired by the way humans learn from others is presented. This method is developed based on cellular learning automata featuring a modular design and cooperation techniques. The modular design brings flexibility, reusability and applicability in a wide range of problems to the method. This paper focuses on analyzing sensitivity of the method's parameters and the applicability in optimization problems. Results of the experiments justify that the new method outperforms similar ones because of employing knowledge sharing technique, reasonable exploration logic, and learning rules based on the action trajectory  

Multi-agent systems (MAS) try to formulate dynamic world which surround human being in every aspect of his life. One of the important challenges encountered in multiagent systems is the credit assignment problem, simply means distributing the result of the work of a group of agents, such that every agent will have the capability of individual learning. This paper presents the result of a solution suggested for multi-agent credit assignment problem. With the help of observing history of credit assignment in the environment, we will understand what actions are reward-deserving. Results are reported on a multi agent domain, addition agents. © 2009 IEEE