Summary
International Symposium on Nonlinear Theory and its Applications
2009
Session Number:B3L-C
Session:
Number:B3L-C1
Common Tensor Discriminant Analysis for EEG Classification
Qibin Zhao, Liqing Zhang, Andrzej Cichocki,
pp.-
Publication Date:2009/10/18
Online ISSN:2188-5079
DOI:10.34385/proc.43.B3L-C1
PDF download (2.2MB)
Summary:
In order to explore underlying multi-mode discriminative information simultaneously from high noise EEG signals, multi-way tensor analysis is more suitable for EEG feature extraction. In this paper, we propose a novel algorithm, called common tensor discriminant analysis (CTDA), to solve the supervised subspace learning by encoding each EEG epoch as a M-order tensor. A tensor-based discriminant analysis framework is presented for simultaneous optimization of a series of projection matrices based on tensor analysis theory and CSP criteria. Furthermore, CTDA has been extended to multiclass case. Experimental results demonstrate the effectiveness and superiority of our proposed algorithms.