Probabilistic methods in BCI research.

This paper suggests a probabilistic treatment of the signal processing part of a brain-computer interface (BCI). We suggest two improvements for BCIs that cannot be obtained easily with other data driven approaches. Simply by using one large joint distribution as a model of the entire signal processing part of the BCI, we can obtain predictions that implicitly weight information according to its certainty. Offline experiments reveal that this results in statistically significant higher bit rates. Probabilistic methods are also very useful to obtain adaptive learning algorithms that can cope with nonstationary problems. An experimental evaluation shows that an adaptive BCI outperforms the equivalent static implementations, even when using only a moderate number of trials. This suggests that adaptive translation algorithms might help in cases where brain dynamics change due to learning effects or fatigue.