Applying Reinforcement Learning to Improve MCOE, an Intelligent Learning Environment for Ecology

Abstract

In the field of Intelligent Learning Environments, the use of multi-agent systems technology is getting more and more attention, and several Machine Learning techniques are being applied on such systems. One of the most promising techniques is Reinforcement Learning. This paper presents the MCOE system, a learning environment for teaching Ecology to children, describing one of its limitations: the low flexibility on the generation of simulated situations for the students. We propose modifications on its architecture to make it more dynamic and adaptable through the generation of a wider set of natural and adaptation-like effects. Instead of building a complex, more deliberative solution, we propose solving the problem using a decentralized and more reactive solution by using reinforcement learning concepts, while keeping the simplicity of the architecture, an important feature in systems designed for small school applications.