Within the area of Brain-Computer Interfacing (BCI), the P300-Speller paradigm, originally proposed by Farwell & Donchin (1988), utilizes the P300 in an oddball-like paradigm to construct a BCI, which is independent of training of participants. In a 6x6 matrix filled with symbols, rows and columns are flashing, resulting in a P300 in the EEG-pattern for an attended symbol.
We fitted the machine-learning technique Support-Vector Machines (SVM) to the demands of the P300-Speller paradigm. We applied the algorithm on data from own experiments and were able to accelerate the P300-Speller paradigm from about 12 bits/min up to 84.7 bits/min, making it a very fast EEG-based BCI. Using this technique, we were able to classify P300 trials with an accuracy of up to 94% within a single trial. By analyzing what was learnt by the SVM, inferences might be made about psychophysiological details.