An adaptive and efficient spatial filter for event-related potentials

A major problem in designing brain computer interface (BCI) systems is the variations of data during long sessions, and also from session to session and subject to subject. This demands extensive training sessions to maintain the overall performance of the systems. As an alternative, here we propose an adaptive version of the πSF method. Taking advantage of simple structure, high performance, and low computational complexity of the algorithm, an extension is developed that can adapt the trained transformation to the new instances. Experimental results confirm that the adaptive method (aπSF) outperforms the non-adaptive versions of πSF and the xDAWN method in an inter-subject train-test schema.