4
$\begingroup$

The SVD and NMF are seem to be very close, so the question: how can I obtain NMF of given matrix from its SVD decomposition?

I've tried to zero-in all negative parts of SVD decomposition, but this gives bad results and iterative approach (zero-in, correct, zero-in) does not help either. Am I missing something or there is no meaningful relation between SVD and NMF?

$\endgroup$
1
  • 1
    $\begingroup$ read the NMF paper papers.nips.cc/paper/… and a course on the SVD (or PCA for the minimization approach), you'll see they are completely different $\endgroup$ – reuns Jul 30 '16 at 15:13
1
$\begingroup$

There is a good reference explaining the relation between PCA and NMF Same author suggested by reuns but with more interpretation

I have found some work done by Chris Ding showing relation between PCA to Clustering. Where PCA is clustering data based on variances while one type of clustering called K-means which cluster the data based on Euclidean Distances. From this point there are proves showing the relation between K-means clustering to NMF.

Brief ppt showing relation between formulas

Good Reference to start with

If the previous one is hard this is having a brief of it

$\endgroup$
3
  • 1
    $\begingroup$ Really nice references. Thank you for sharing. $\endgroup$ – Royi Oct 2 '17 at 16:58
  • 1
    $\begingroup$ Could you please update second reference (the link is broken). $\endgroup$ – Moonwalker Oct 5 '17 at 21:51
  • 1
    $\begingroup$ @Moonwalker done $\endgroup$ – Mour_Ka Oct 5 '17 at 22:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.