Fast Algorithms for Sparse Reduced-Rank Regression - École des Ponts ParisTech Access content directly
Journal Articles Proceedings of Machine Learning Research Year : 2019

Fast Algorithms for Sparse Reduced-Rank Regression

Abstract

We consider a reformulation of Reduced-Rank Regression (RRR) and Sparse Reduced-Rank Regression (SRRR) as a non-convex non-differentiable function of a single of the two matrices usually introduced to parametrize low-rank matrix learning problems. We study the behavior of proximal gradient algorithms for the minimization of the objective. In particular, based on an analysis of the geometry of the problem, we establish that a proximal Polyak-Łojasiewicz inequality is satisfied in a neighborhood of the set of optima under a condition on the regularization parameter. We consequently derive linear convergence rates for the proximal gradient descent with line search and for related algorithms in a neighborhood of the optima. Our experiments show that our formulation leads to much faster learning algorithms for RRR and especially for SRRR.
Fichier principal
Vignette du fichier
865.pdf (1.02 Mo) Télécharger le fichier
865-supp.pdf (719.37 Ko) Télécharger le fichier
Origin Files produced by the author(s)
Loading...

Dates and versions

hal-02075623 , version 1 (22-03-2019)

Identifiers

  • HAL Id : hal-02075623 , version 1

Cite

Benjamin Dubois, Jean-François Delmas, Guillaume Obozinski. Fast Algorithms for Sparse Reduced-Rank Regression. Proceedings of Machine Learning Research, 2019, Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, 89, pp.2415-2424. ⟨hal-02075623⟩
261 View
336 Download

Share

Gmail Mastodon Facebook X LinkedIn More