I am trying to implement Normalized Cut Algorithm . In the original paper (Shi & Malik), they compute the k smallest eigenvectors of:
(D-W)x = lambda Dx
which is equivalent to solving: D^(-1/2) (D-W) D^(-1/2) x = lambda x
D being a diagonal matrix and W a sparse symmetric matrix.
Is there a way to modify the algorithm or the eigensystem to compute the k largest eigenvectors instead and still find the normalized cut ?
Thanks