Unsupervised adaptation of the LDA classifier for Brain-Computer Interfaces.
Carmen Vidaurre, Alois Schloegl, Benjamin Blankertz, Motoaki Kawanabe and Klaus-Robert Müller
Proceedings of the 4th International Brain-Computer Interface Workshop and Training Course 2008
This paper discusses simulated on-line unsupervised adaptation of the LDA classier in
order to counteract the harmful eect of non-class related non-stationarities in EEG during
BCI sessions. Three types of adaptation procedures were applied to the two large BCI data
sets from TU Graz and Berlin BCI project. Our results demonstrate that the unsupervised
adaptive classiers can improve performance substantially under dierent BCI settings. More
importantly, since label information is not necessary, they are applicable to wide ranges of
practical BCI tasks.