POSTECH BCIs with machine learning

This paper outlines a brief overview of brain computer interfaces (BCIs), the research on which has been conducted at POSTECH machine learning lab. It has three folds. First, matrix factorization methods are introduced, which are used to learn spectral features for automatic classification of brain waves. Second, Bayesian multi-task learning methods are presented, which are applied to multi-subject EEG classification where subject-to-subject transfer is often considered to improve EEG classification. Third, tongue-machine interface is presented, where glossokinetic potentials involving tongue movements are analyzed to predict where tongue touches around gum line.