Re: what is SVM? what are the good features of SVM?

"Ted Dunning" <[EMAIL PROTECTED]> wrote:
> Interestingly, regularized logistic regression is very nearly
> equivalent to SVM techniques and can be trained more efficiently
> the normal QP algorithms used for training SVM's.  Adding the
> regularization also greatly improves the convergence properties of
> logistic regression itself.

Thanks for the links. Yes, in my experience SVM's with the LINEAR
KERNEL are very similar to logistic regression. But the criteria are
different: linear SVM with its separability criterion is better for
classification, while logistic regression works better for modelling
the outcome/label/class probability density in uncertain domains.
Furthermore, SVM's often benefit from regularization, so we can only
talk about the similarity between regularized linear SVM and
regularized logistic regression.

But the key idea of SVM is the arbitrariness of the kernel. SVM with
an RBF or a polynomial kernel has very little in common with logistic
regression.

> I haven't seen anybody present the relatively simple proof that
> demonstrates the near equivalence of regularized logistic regression
> and structural risk minimization.  It is simple enough that somebody
> must have published it already.

I'm working on a paper where I represent linear SVM as a logistic
regression model. The models may take an identical form, but both the
optimality criteria and fitting methods are different.

Aleks