Randomized pick-freeze for sparse Sobol indices estimation in high dimension

Castro, Yohann de¹  ; Janon, Alexandre² 

¹ UniversitéParis-Sud, Laboratoire de Mathématiques d’Orsay, Bâtiment 425, Université Paris-Sud, 91405 Orsay, France
² UniversitéParis-Sud, Laboratoire de Mathématiques d’Orsay, Bâtiment 425, Université Paris-Sud, 91405 Orsay, France

ESAIM: Probability and Statistics, Tome 19 (2015), pp. 725-745

Résumé

This article investigates selection of variables in high-dimension from a non-parametric regression model. In many concrete situations, we are concerned with estimating a non-parametric regression function $f$ that may depend on a large number $p$ of inputs variables. Unlike standard procedures, we do not assume that $f$ belongs to a class of regular functions (Hölder, Sobolev, ...), yet we assume that $f$ is a square-integrable function with respect to a known product measure. Furthermore, observe that, in some situations, only a small number $s$ of the coordinates actually affects $f$ in an additive manner. In this context, we prove that, with only $𝒪\left(slogp\right)$ random evaluations of $f$, one can find which are the relevant input variables with overwhelming probability. Our proposed method is an unconstrained $\ell {}_1$-minimization procedure based on the Sobol’s method. One step of this procedure relies on support recovery using $\ell {}_1$-minimization and thresholding. More precisely, we use a thresholded-LASSO to faithfully uncover the significant input variables. In this frame, we prove that one can relax the mutual incoherence property (known to require $𝒪\left(s^{2}logp\right)$ observations) and still ensure faithful recovery from $𝒪\left(s^{\alpha }logp\right)$ observations for any $1\le \alpha \le 2$.

Détail

DOI : 10.1051/ps/2015013

Classification : 62G08, 62G35, 65H10, 93A30, 93B35
Keywords: Sensitivity analysis, Sobol indices, high-dimensional statistics, LASSO, Monte-Carlo method