The two improved algorithms of ICA
Yunjie Zhang,Zhao, Feng,Min Cai
DOI: https://doi.org/10.1109/ICEBEG.2011.5882745
2011-01-01
Abstract:The FastICA algorithm and the natural gradient algorithm is widely used in blind signal separation, as the two more popular algorithms in the areas of the ICA, because they can find implicit in the observation data in the independent component. However, they each have drawback, such as, online data processing and mass data(such as image data)processing tasks, we need that FastICA algorithm is faster and more efficient; the same time we also hope that the nature gradient algorithm can not only deal with symmetric distribution of the signal, but also asymmetrical distribution of the signal. Therefore, this paper establishes a fifth-order convergence of the Newton iterative algorithm as a basis for improvement of the FastICA algorithm; farther more the paper also gives an improved natural gradient method which is used for symmetrical distribution of the signal and asymmetrical distribution of the signal.