V-fold cross-validation improved: V-fold penalization

Sylvain Arlot

We study the efficiency of V-fold cross-validation (VFCV) for model selection from the non-asymptotic viewpoint, and suggest an improvement on it, which we call ``V-fold penalizati…