Most independent component analysis (ICA) algorithms use mutual information (MI) measures based on Shannon entropy as a cost function, but Shannon entropy is not the only measure in the literature. In this paper, instead of Shannon entropy, Tsallis entropy is used and a novel ICA algorithm, which uses kernel density estimation (KDE) for estimation of source distributions, is proposed. KDE is directly evaluated from the original data samples, so it solves the important problem in ICA: how to choose nonlinear functions as the probability density function (pdf) estimation of the sources.

Most independent component analysis (ICA) algorithms use mutual information (MI) measures based on Shannon entropy as a cost function, but Shannon entropy is not the only measure in the literature. In this paper, instead of Shannon entropy, Tsallis entropy is used and a novel ICA algorithm, which uses kernel density estimation (KDE) for estimation of source distributions, is proposed. KDE is directly evaluated from the original data samples, so it solves the important problem in ICA: how to choose nonlinear functions as the probability density function (pdf) estimation of the sources.