Whitening or sphereing¶
Is a process that ensures that the empirical covariance matrix is proportional to \(I\), so the data is uncorrelated and of equal variance along each dimension. For each \(x\) we compute:
\[\Lambda^{-\frac{1}{2}}U^Tx\]
\(\Lambda\) matrix of eigenvalues of X
\(U\) the eigen vectors of \(\Sigma\)