A Bayesian approach for adaptive BCI classification

In this article, we present an adaptive classifier for BCI based on a mixture of Gaussian (moG) model of the features and a dynamical Bayesian model of the class means. We apply this approach to feedback data from the Berlin Brain- Computer Interface (BBCI). The proposed model can improve the classification performance by compensating for substantial changes of EEG signals between training and feedback sessions as well as for gradual nonstationarity in the feedback sessions.