Paper on SVMs

[Robert Urbanczik]

Dear Connectionists,

the following paper (5 pages, to appear in Phys. Rev. Letts.) is 
available from:

ftp://ftp.physik.uni-wuerzburg.de/pub/preprint/2001/WUE-ITP-2001-006.ps.gz
   

               
               M. Opper and R. Urbanczik 

     Universal learning curves of support vector machines 

ABSTRACT:
  Using methods of Statistical Physics, we investigate the r\^ole of
  model complexity in learning with support vector machines (SVMs),
  which are an important alternative to neural networks.  We show the
  advantages of using SVMs with kernels of infinite complexity on noisy
  target rules, which, in contrast to common theoretical beliefs, are
  found to achieve optimal generalization error although the training
  error does not converge to the generalization error.  Moreover, we
  find a universal asymptotics of the learning curves which only depend
  on the target rule but not on the SVM kernel.

_________________________________________________________________________

R. Urbanczik                         Email: 
Inst. for Theoretical Physics III       urbanczik at physik.uni-wuerzburg.de
University Wuerzburg                 Phone: 
Am Hubland                               ++49 931 888 4908
97074 Wuerzburg                      Fax:   
Germany                                  ++49 931 888 5141
_________________________________________________________________________