High-order contrasts for independent component analysis
Neural Computation
ACM Computing Surveys (CSUR)
Self-Organising Neural Networks: Independent Component Analysis and Blind Source Separation
Self-Organising Neural Networks: Independent Component Analysis and Blind Source Separation
Hi-index | 0.00 |
How to extract task-relevant components from spontaneous electroencephalogram background is an open problem in EEG signal analysis. An Independent Component Automatic Clustering (ICAC) method, which combined Independent Component Analysis (ICA) with automatic clustering, is developed in this paper. In ICAC, the ICA decomposed components were grouped into several clusters and sorted automatically. A majority of task-relevant components could be grouped into one cluster and be recognized easily, which can compensate the traditional ICA limitation of component sorting without any task specialized orders. We adopted this method on multi trails EEG signals during imaginary hand movement, results showed that ICAC can automatically extract task-relevant component and increase the Fisher Criterion (FC) separability significantly. Furthermore, we show that the residual mutual information between task-relevant components is not useless as previously regarded but very useful on components recognition.