Specialized Multi-Agents Learning System

Speaker: Nasser Mooman

With the increasing amount of available information sources for learners, the need for systems that can effectively and efficiently mine, retrieve, and process such information has become critical. Traditional approaches of mining and retrieving the desired information need to scale with the growth of information by mapping the required information to the users as well as building learning models. We propose a novel approach that provides an intelligent learning mechanism to learn users needs in a large dynamic information space by the use of specialized learning agent paradigm. Artificial Intelligent (AI) techniques such as multi-agents, reinforcement learning, and data mining are appropriate approaches to build a system that assists the users within a specific domain by learning the users' preference and behaviors.

The framework of the specialized multi-agent system will be addressed in this presentation. Reinforcement Learning (RL) will be discussed as an advanced design tool that enables the construction of intelligent learning agents. Specifically we employ two types of RL algorithms in our study , Q-learning and SARSA algorithms. The presentation also will address the multi-agent collaboration method that is proposed to communicate between the learning agents.