作者: N. Jeremy Hill , Thomas Navin Lal , Michael Schröder , Thilo Hinterberger , Guido Widman
DOI: 10.1007/11861898_41
关键词: Unsupervised learning 、 Independent component analysis 、 Speech recognition 、 Overfitting 、 Electroencephalography 、 Pattern recognition (psychology) 、 Computer science
摘要: We employed three different brain signal recording methods to perform Brain-Computer Interface studies on untrained subjects. In all cases, we aim develop a system that could be used for fast, reliable preliminary screening in clinical BCI application, and are interested knowing how long sessions need be. Good performance achieved, average, after the first 200 trials EEG, 75–100 MEG, or 25–50 ECoG. compare of Independent Component Analysis Common Spatial Pattern algorithm each sensor types, finding spatial filtering does not help helps little ECoG, improves great deal EEG. cases unsupervised ICA performed at least as well supervised CSP algorithm, which can suffer from poor generalization due overfitting, particularly ECoG MEG.