iEnsemble: A Framework for Committee Machine Based on Multiagent Systems with Reinforcement Learning

The Machine Learning is one of the areas of Artificial Intelligence whose objective is the development of computational techniques for knowledge and building systems able to acquire knowledge automatically. One of the main challenges of learning algorithms is to maximize generalization. Thus the board machine, or a combination of more of a learning machine approach known in literature with the denomination ensemble along with the theory agents, become a promising alternative in this challenge. In this context, this research proposes the iEnsemble framework, which aims to provide a model of the ensemble through a multi-agent system architecture, where generalization, combination and learning are made through agents, through the performance of their respective roles. In the proposal, the agents follow each their life cycle and also perform the iStacking algorithm. This algorithm is based on Stacking method, which uses the reinforcement learning to define the result of the Ensemble. To validate the initial proposal of the framework, some experiments have been performed and the results obtained and limitations are presented.