Cross-validation for densities and regressions

Abstract

Virtually all nonparametric (infinite dimensional) problems require the choice of a smoothing parameter. It is natural to attempt to use the data to guide the choice of this parameter. The most widely applicable data driven technique is cross validation. The method of cross validation is introduced and results are presented which establish consistency for certain cross validated density estimators and consistency as well as asymptotic normality for rigid regression.