My understanding is that Bayesian classifiers should have comparable
or better generalization performance to SVM.
Main advantage of SVM would be speed, when compared to Bayesian classifiers
since SVM approach boils down to minimization of quadratic function with
linear constraints whereas in general Bayesian classification approach it is
necessary to
execute complex multidimensional integrals, which have to be approximated
either by integrals
over gaussian distributions or calculated numerically using e.g Monte Carlo
approach.
It would be interesting to hear an opinion of people who have more
experience in this area.
R
Ryszard Czerminski phone: (781)994-0479
ArQule, Inc. e-mail: [log in to unmask]
19 Presidential Way http://www.arqule.com <http://www.arqule.com/>
Woburn, MA 01801 fax: (781)994-0679
-----Original Message-----
From: SatPrem Reddy [mailto:[log in to unmask]]
Sent: Tuesday, October 23, 2001 2:15 PM
To: [log in to unmask]
Subject: generalization of svms
Hi all,
Why is it that SVMs have good generalization performance
compared to other classifiers? I have read that it is because
of Structural Risk Minimization. But, don't Bayesian classifiers
also use SRM? What is it that really sets SVM apart from
others?
TIA.
-SatPrem
|