Ensemble Machine-Learning Applied to Brain-Computer Interfaces
In machine learning, the expansive use of Artificial Neural Networks (ANNs) in different areas can be justified by its succesfull results achieved in problems that before were unsolveable with classic programming, which, in a certain sense, required some interaction with experts on the field. The huge potential of the ANNs can be attributed to its plasticity and nonlinearity, allowing its adaptation to many contexts and different levels of complexity.
The wide range of problems able to be addressed with machine learning techniques contributed to the emergence of several ANN structures endowed with distinct aspects, which contributed to its use for different types of data. However, in order to deal with increasingly complex problems, there emerged the necessity of combining the power of action of more than one type of ANN, giving rise to the Committee Machines or Ensembles. In such case, each machine in the ensemble acts like an expert whose output is combined with that of the other experts in order to reach a consensus, allowing the achievement of improved performance. This approach reveals a horizon of possible new applications to be explored. This is the case, for instance, of the Brain-Computer Interfaces, which aims to take actions from commands triggered by thought patterns. The variability of thought patterns for a single user as well as the variability observed among different users makes the idea of a multiuser system very challenging. Based on this, the present work aims at using Committee Machines (composed by several artificial neural networks with different structures) to obtain a robust multiuser BCI system. The results might lead to a new understanding of the relation of thought patterns amongst distinct users.