A tensorial approach to single trial recognition for Brain Computer Interface

In this paper, we propose a tensorial approach to single trial recognition in a EEG-based BCI system related to movement related potentials. In this approach input data are considered as tensors instead of more conventional vector or matrix representations. Feature extraction for multiway EEG spectral tensors is solved by using tensor (multi-array) decompositions. For the same EEG motor imagery dataset, the developed algorithms improved the accuracy of classification by almost 10% compared with the common spatial pattern (CSP) method.