Weighted covariance methods

The eigenvalue decomposition of the weighted covariance matrix data, as explained in Section 4.3.6, allows the computation of the ICA estimates using standard methods of linear algebra [19] on matrices of reasonable complexity ($m\times m$). Here, the data must be sphered. This method is computationally highly efficient, but, unfortunately, it works only under the rather severe restriction that the kurtoses of the independent components are all different.