Some theoretical results on the grouped variables Lasso
Résumé
We consider the linear regression problem with Gaussian error. We estimate the unknown parameters via an estimator constructed from a grouped variables penalty. It can be viewed as a slight modification of the Group Lasso estimator introduced by Yuan and Lin (2006). We establish several new theoretical results which prove that the considered estimator exploits more the sparsity in the model than the well-known Lasso estimator.
Domaines
Statistiques [math.ST]
Origine : Fichiers produits par l'(les) auteur(s)