1
$\begingroup$

Is there a general way of fitting a gaussian mixture with $m$ components with a gaussian mixture with $n \ll m$ components, short of generating samples and redoing a full fit? I expect a lot of components to be "naturally" merge-able, i.e. with "close" means and variances.

Thanks.

$\endgroup$
1
$\begingroup$

This paper might be something in that direction.

I hope this helps.

(can't add this as a comment as I don't have enough reputation.)

$\endgroup$
  • $\begingroup$ Please add this as a comment, not as an answer. An answer should contain a brief summary of the paper or the idea at least. $\endgroup$ – Mayou36 Jul 23 '17 at 10:51
  • $\begingroup$ @Mayou36, unfortunately I can't as I don't have enough points. $\endgroup$ – DataD'oh Jul 23 '17 at 12:25
  • $\begingroup$ okey then. May add some comments to what the paper is about? Or a short summary? $\endgroup$ – Mayou36 Jul 23 '17 at 15:49

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.