Representations of directions in EEG-BMI using winner-take-all readouts

Abstract

In EEG-BMI systems, how to represent user's intention is a most important question. The motor imagery method has used to represent directions where user want machine to move. However, the motor imagery method is just mapping the parts of bodies to directions such as a left hand means moving left. We study novel methods for representations of directions not using the motor imagery. First, we record the EEG signals when a user thought direction where want to move. Second, we used echo state networks paradigm which is one of Reservoir computing method for analysis and classification of non-linear time series. Third, we designed winner-take-all readouts for representations of user's intended directions. These winner-take-all readouts are perfectly classified directions of user's intention using EEG signals.