If I use a manifold learning method to project some data points into a low dimensional space, what will be the distances between the projected points? Can I use Euclidean distance? If the distances will not be Euclidean, what kind of distance metric do I have to use and how?


It would depend on the manifold learning technique, I think.

DrLIM, for instance is Euclidean, but I've seen other metrics. Section 6 of this paper used cosine distance, and I've seen Hamming distance as well.

Which method did you have in mind?

  • $\begingroup$ I wanted to use aaai.org/ocs/index.php/AAAI/AAAI10/paper/view/1924. $\endgroup$ – Hossein Abedi Apr 27 '13 at 19:05
  • $\begingroup$ @Hossein Equation 2 seems to suggest that they're using Euclidean distance, but I'm not sure. $\endgroup$ – David J. Harris Apr 27 '13 at 19:19
  • $\begingroup$ That norm is because of this : "Each point of an n-dimensional manifold has a neighbourhood that is homeomorphic to the Euclidean space of dimension n.", -from wikipedia! $\endgroup$ – Hossein Abedi Apr 27 '13 at 19:39

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.