Eriksson, J., Karvanen, J., Koivunen, V.
Source Distribution Adaptive Maximum Likelihood
Estimation of ICA Model
In this paper a new approach for performing Independent Component
Analysis (ICA) is introduced. The Extended Generalized Lambda
Distribution (EGLD) is employed for modeling source distributions. The
major benefit of the EGLD is that it also takes into account the
skewness of the distributions. We briefly review maximum likelihood
approach in ICA and study how the parameters of EGLD may be estimated.
The score function of EGLD based ICA is presented and algorithms for
its maximization are proposed. The simulation examples illustrate that
the proposed method reliably separates the sources in situations where
some widely used contrast functions may perform poorly.