In Vapnik's book "Statistical Learning Theory", Theorem 10.5 states that - for a Support Vector Machine - the expected probability of error (of the optimal hyperplane) is upper bounded by $1/(l+1)$ times the expected number of support vectors.

Is the expected probability of error - the LHS of inequality 10.32 - non-increasing in the number $l$ of training samples?

  • 3
    I do not see why your first paragraph does not answer your question. Can you make the question stand on its own, so that it can be understood without referring to the book? – Sasho Nikolov Jan 20 '14 at 14:39
  • Sasho, sorry but I do not understand what do you mean with your first comment. Do you mean that the claim is true? – user693 Jan 20 '14 at 16:35
  • Disregard the first comment, I misunderstood. Still, could you reword the question so that it can be understood without looking up an equation in the book? – Sasho Nikolov Jan 20 '14 at 17:14
  • 3
    Why does my answer to cstheory.stackexchange.com/questions/5314/… not answer your question? – Lev Reyzin Feb 7 '14 at 22:32
  • 1
    When I first read the title, I thought it was some kind of joke about the number of errors in the book ! – Suresh Venkat Feb 7 '14 at 23:42

Your Answer

 
discard

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.