AN ITERATIVE NONLINEAR GAUSSIANIZATION ALGORITHM FOR RESAMPING DEPENDENT COMPONENTS

Jen-Jen Lin 
emails: jjlin@mcu.edu.tw, saito@math.ucdavis.edu, levine@wald.ucdavis.edu

We propose an Iterative Nonlinear Gaussianization Algo- rithm (INGA), which seeks a nonlinear map from a set of dependent random variables to independent Gaussian ran- dom variables. A direct motivation of the INGA is to extend the principal component analysis (PCA) which transforms a set of correlated random variables into uncorrelated (inde- pendent up to second order) random variables. An obvious advantage of deriving independent components is that we can simulate a stochastic process of dependent multivari- ate variables by sampling univariate independent variables. The quality of the transformation is evaluated by statis- tical tests on the Kullback-Leibler (KL) distance between the distribution of the transformed variables the standard multivariate Gaussian distribution N(0; I). The quality of the simulations is evaluated quantitatively by the statistics of the KL distances between the sample mean distribution of the original samples and that of the simulated samples. Several numerical examples including synthetic and real- life image databases show the capabilities and limitations of INGA.