Nonlinear Acceleration of CNNs - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2018

Nonlinear Acceleration of CNNs

Accélération Nonlinéaire des réseaux de neurones convolutionnels.

Résumé

The Regularized Nonlinear Acceleration (RNA) algorithm is an acceleration method capable of improving the rate of convergence of many optimization schemes such as gradient descend, SAGA or SVRG. Until now, its analysis is limited to convex problems, but empirical observations shows that RNA may be extended to wider settings. In this paper, we investigate further the benefits of RNA when applied to neural networks, in particular for the task of image recognition on CIFAR10 and ImageNet. With very few modifications of exiting frameworks, RNA improves slightly the optimization process of CNNs, after training.
Fichier principal
Vignette du fichier
ICLR2018_Workshop.pdf (172.39 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01805251 , version 1 (01-06-2018)
hal-01805251 , version 2 (01-06-2018)
hal-01805251 , version 3 (01-06-2018)

Identifiants

  • HAL Id : hal-01805251 , version 3

Citer

Damien Scieur, Edouard Oyallon, Alexandre d'Aspremont, Francis Bach. Nonlinear Acceleration of CNNs. ICLR Workshop track, Apr 2018, Vancouver, Canada. ⟨hal-01805251v3⟩
305 Consultations
89 Téléchargements

Partager

Gmail Facebook X LinkedIn More