A Fast Accurate Training-less P300 Speller: Unsupervised Learning Uncovers new Possibilities

The P300 speller paradigm is a Brain-Computer Interface, introduced in 1988 by Farwell and Donchin, which allows the users to type by simply focusing their attention on the desired symbol in a spelling matrix. This type of BCI has already been applied in a real world setting for patients, but is severely limited by the need for a tedious calibration session before each usage, whereby the system is trained supervisedly. In this demonstration we present our solution to this calibration session: an unsupervised P300 speller. The system is able to quickly learn online, entirely without labeled data, thus bypassing the need for calibration. The spelling accuracies are comparable to supervised spellers.