On the need for alternative feedback training approaches for BCI

One of the most serious issue with current BCI systems is their lack of reliability and poor performances. These poor performances are due in part to the imperfect signal processing algorithms used, which are not yet able to extract robustly a relevant information from EEG signals despite the various noise sources, the signal non-stationarity and the limited amount of data available. However, this is most probably not the only reason that can explain such poor performance and reliability. In particular, there are several other components of the BCI loop that may also be deficient. This includes, for instance, the interaction technique used, which has to be carefully designed and improved as well, but this can also be the user himself who may not be able to produce reliable EEG patterns. If this is the case, whatever the signal processing algorithms used, there would be no way to properly identify the mental command produced by the user. Despite this, the BCI community has focused the majority of its research efforts on signal processing and machine learning, mostly neglecting the human in the loop.