Semi-supervised temporal-spatial filter based on MRP for brain-computer interfaces

In brain-computer interface (BCI) studies, if the number of training trails is small, the discriminative patterns of movement related potentials (MRPs) can not be appropriately extracted by temporal-spatial filter (TSF) algorithm. Thus in this paper, we proposed a semi-supervised TSF (ssTSF) algorithm which employed self-training scheme to induce the unlabelled trails with high confidences and learn the discriminative patterns of MRPs iteratively. We compared TSF and ssTSF algorithm on the data from BCI competition I. The results demonstrated the effectiveness of the ssTSF, especially for small training sets.