This answer to Data normalization for RBF kernel points out that RBF kernel implies Eucledean distance. Are there kernels corresponding to other popular distance/dissimilarity measures, such as Bray-Curtis or Jensen-Shannon? Or could they be easily designed? What are the constraints to take into account (e.g., does Bray-Curtis pose a problem because of its non-analiticity?)

Related: About Gaussian kernel for distances other than Euclidian

  • $\begingroup$ Bray-Curtis dissimilarity isn't a distance because it doesn't satisfy the triangle inequality. In any event, the thing that you care about for the purposes of SVM is that the kernel matrix is symmetric and positive definite for distinct inputs. $\endgroup$
    – Sycorax
    Sep 27 at 13:30
  • $\begingroup$ @Sycorax I corrected the question accordingly. If using a real distance is a constraint for kernel trick, this could be the beginning of an answer to the last part of the question. $\endgroup$
    – Roger V.
    Sep 27 at 13:43


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Browse other questions tagged or ask your own question.