Low-Dimensional Subject Representation-Based Transfer Learning in EEG Decoding

Recently, the advances in passive brain-computer interfaces (BCIs) based on electroencephalogram (EEG) have shed light real-world neuromonitoring technologies. However, human variability EEG activities hinders development of practical applications EEG-based BCI. To tackle this problem, many transfer-learning techniques perform supervised calibration. This kind calibration approach requires task-relevant data, which is impractical real-life scenarios such as drowsiness during driving. study presents a framework for decoding low-dimensional representations subjects learned from pre-trial EEG. Tensor decomposition was applied to extract underlying characteristics subject, spatial, and spectral domains. Then, proposed assessed obtain subject that with similar brain dynamics can be identified. method leverage existing data other users, small number rapid, non-task, unsupervised new user build an accurate Our results demonstrated that, terms prediction accuracy, representation-based transfer learning (LDSR-TL) outperformed random selection, Riemannian manifold cognitive-state tracking, while requiring fewer training data. The greatly improve practicability, usability BCI real world.