Natural Regularization in SVMs

Recently the so called Fisher kernel was proposed by to con struct discriminative kernel techniques by using generative models We provide a regularization theoretic analysis of this approach and extend the set of kernels to a class of natural kernels all based on generative models with density p xj like the original Fisher ker nel This allows us to incorporate distribution dependent smooth ness criteria in a general way As a result of this analyis we show that the Fisher kernel cor responds to a L p norm regularization Moreover it allows us to derive explicit representations of the eigensystem of the kernel give an analysis of the spectrum of the integral operator and give experimental evidence that this may be used for model selection purposes