Amaresh V Malipatil wrote:
>
> Hello,
> Could anyone throw light upon, how do we go about choosing a kernel function for
> non-linear mapping. Is it intuitive or is it possible to precisely choose a
> function given a problem at hand.
>
> thanks,
> --
> Amaresh V Malipatil
> Graduate Student
> Department of Electrical Engineering
> University of Notre Dame
> Indiana 46556
Hi Amaresh
Either polynomial/rbf kernel will be suitabe for nonlinear problem with
SVM. We did some research about automatic kernel selection for SVM based
on meta-learning approach. The research has been submited in IEEE
Transactions.That is, from a wide data characteristics we can say which
kernel is more suitable for your problem. You can think in different way
also. Still it is a hot research issue for SVM researcher.
Cordial regards.
S.Ali
--------------------------------
Shawkat Ali
School of BusSys
Faculty of IT
Monash University
VIC 3800,Australia
--------------------------------------------
Mobile : 0403 613 791 (AU)
61 403 613 791 (OS)
Phone : 61 03 9905 5353 (w)
Fax : 61 03 9905 5159
E-mail : [log in to unmask]
Home page: http://www.bsys.monash.edu.au/people/shawkat/
|