Anticorrelated Noise Injection for Improved Generalization - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Anticorrelated Noise Injection for Improved Generalization

Résumé

Injecting artificial noise into gradient descent (GD) is commonly employed to improve the performance of machine learning models. Usually, uncorrelated noise is used in such perturbed gradient descent (PGD) methods. It is, however, not known if this is optimal or whether other types of noise could provide better generalization performance. In this paper, we zoom in on the problem of correlating the perturbations of consecutive PGD steps. We consider a variety of objective functions for which we find that GD with anticorrelated perturbations ("Anti-PGD") generalizes significantly better than GD and standard (uncorrelated) PGD. To support these experimental findings, we also derive a theoretical analysis that demonstrates that Anti-PGD moves to wider minima, while GD and PGD remain stuck in suboptimal regions or even diverge. This new connection between anticorrelated noise and generalization opens the field to novel ways to exploit noise for training machine learning models.
Fichier principal
Vignette du fichier
orvieto22a.pdf (3.83 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03883872 , version 1 (04-12-2022)

Identifiants

  • HAL Id : hal-03883872 , version 1

Citer

Antonio Orvieto, Hans Kersting, Frank Proske, Francis Bach, Aurelien Lucchi. Anticorrelated Noise Injection for Improved Generalization. ICML 2022 - 39th International Conference on Machine Learning, Jul 2022, Baltimore, United States. ⟨hal-03883872⟩
14 Consultations
19 Téléchargements

Partager

Gmail Facebook X LinkedIn More