Abstract: The Independent Component Analysis(ICA) is a recently developed method for multi-signal processing and Blind Source Separation(BSS). However, its constraint on the sources that the sources are statistically independent of each other greatly limits its applications to BSS since the sources in most applications are not guaranteed to be independent. This paper presents a partially independent component analysis(PICA) method for BSS of dependent sources, where the approximately independent indices of the sources are selected with some feature selection method, and ICA is performed on the selected indices of the observations. A large number of simulations and a real world DNA microarray data experiment show great availability and effectiveness of the method presented here.

Key words: independent component analysis, PICA, feature selection, blind source separation