Sélection de variables pour l'apprentissage simultanée de tâches

Rémi Flamary (1) , Alain Rakotomamonjy (1) , Gilles Gasso (1) , Stephane Canu (1)

1 LITIS - Laboratoire d'Informatique, de Traitement de l'Information et des Systèmes

Recently, there has been a lot of interest around multi-task learning (MTL) problem with the constraints that tasks should share common features. Such a problem can be addressed through a regularization framework where the regularizer induces a joint-sparsity pattern between task decision functions. We follow this principled framework but instead we focus on lp − l2 (with p ≤ 1) mixed-norms as sparsity-inducing penalties. After having shown that the l1 − l2 MTL problem is a general case of Multiple Kernel Learning (MKL), we adapted the available efficient tools of solving MKL to the sparse MTL problem. Then, for the more general case when p < 1, the use of a DC program provides an iterative scheme solving at each iteration a weighted 1 − 2 sparse MTL problem.

Mots clés: Apprentissage Multi-tâches; Sélection de variable; Méthodes à noyaux

Domaines: Apprentissage [cs.LG]

Rémi Flamary, Alain Rakotomamonjy, Gilles Gasso, Stephane Canu. Sélection de variables pour l'apprentissage simultanée de tâches. Confrénce D'Apprentissage (CAp), May 2009, Hammamet, France. pp.109-120. ⟨hal-00452332⟩