I have a question about the support vector machine (SVM) kernel trick. How do you find the boundaries of the training data set in kernel projected space? Is that the same boundaries as you can obtain in original input space? Does kernel mapping change the data distribution? I am new to SVM, can someone please give me a straightforward explanation or links to related information?
Asked
Active
Viewed 225 times
1
-
boundaries in input space can be find via convex-hull or concaved-hull yet in kernel space there is no obvious way to find such exact boundaries as geometrical algorithms are not feasible. – tommi Sep 27 '13 at 19:15
-
see http://stats.stackexchange.com/questions/168051/kernel-svm-i-want-an-intuitive-understanding-of-mapping-to-a-higher-dimensional/168082#168082 – Aug 22 '15 at 07:44