Brain state differences between calibration and application session influence BCI classification accuracy
Matthias Krauledat, Florian Losch and Gabriel Curio
Proceedings of the 3rd International Brain-Computer Interface Workshop and Training Course 2006
The Berlin Brain-Computer Interface (BBCI) has been developed to
transfer the main load of learning from the user to the machine.
After a short calibration period of approx. 30 minutes, even
untrained users with no previous BCI experience can achieve
bit-rates of more than 35 bits/min. In some of these experiments,
however, the classifier from the calibration period needs to be
slightly adapted by adding a constant bias term to its output in
order to maintain a stable performance throughout the feedback
session. In this paper, we will provide evidence that a change in
the brain states between calibration and feedback periods probably
causes this need for adaptation.