Regularization method for ICA with reference
Liao Gui-sheng · Computer Engineering and Applications Journal · 2009
Independent Component Analysis with Reference(ICA-R) utilizes a priori information or reference signal and achieves good separation results,but its threshold parameter is very hard to determine and its computation load is very great.Theoretic analysis and experiments shows ICA-R even can't converge if the threshold is improperly selected.By inserting the closeness measure function of ICA-R as a regularization term into the usual negentropy contrast function for FastICA,a very simple improved algorithm is proposed.Experiments with synthetic signals,real ECG data demonstrate its quick convergence,good separation and flexible selection for the regularization parameter as well as.