Call for NIPS 2008 Kernel Learning Workshop Submissions
Document Actions
N. Oliver, B. Sch\"olkopf, and A. J Smola (2000)
Natural Regularization from Generative Models
In: Advances in Large Margin Classifiers, ed. by A.J. Smola and P.L. Bartlett and B. Schölkopf and D. Schuurmans, pp. 51-60, Cambridge, MA, MIT Press.
A class of kernels including the Fischer kernel recently proposed by Jaakola and Haussler is analyzed. The analysis hinges on information-geometric properties of the log probability density function (generative model) and known connections between support vector machines and regularization theory, and proves that the maximal margin term induced by the considered kernel corresponds to a penalizer computing the $L_2$ norm weighted by the generative model. Moreover, it is shown that the feature map corresponding to the kernel whitens the data.