by Ingo Steinwart
Preprint series: 02-03, Reports on Analysis and Computer Science
The paper is published: Jenaer Schriften zur Mathematik und Informatik, Math/Inf/09/02
Abstract: We show that various classifiers that are based on a minimization of a regularized risk are universally
consistent. In particular, several types of support vector machines as well as regularization networks are
treated. Our methods combine techniques from stochastic, approximation theory and functional analysis.
Keywords: Computational Learning Theory, Pattern Recognition, PAC Model, Support Vector Machines, Kernel Methods, Universal Consistency