Linear and nonlinear ICA based on mutual information

Independent component analysis (ICA), both linear and nonlinear, one of the best methods is minimum the mutual information (MI) of the estimated components. Sometimes it is exist many local minima, especially nonlinear mixtures. genetic algorithm (GA) is a method against local minima, and have a high degree of flexibility in the evaluation function. This paper introduce the MI theories, and use MA and GA into the linear and nonlinear ICA. The method of adaptive algorithms for ICA will be helpful to further study, the last give some experimental results.