« Problème de l'explosion du gradient » : différence entre les versions
Aucun résumé des modifications |
Aucun résumé des modifications |
||
| Ligne 11 : | Ligne 11 : | ||
== Termes privilégiés == | == Termes privilégiés == | ||
=== === | |||
| Ligne 18 : | Ligne 18 : | ||
===Exploding Gradient Problem=== | |||
The Exploding Gradient Problem is the opposite of the Vanishing Gradient Problem. In Deep Neural Networks gradients may explode during backpropagation, resulting number overflows. A common technique to deal with exploding gradients is to perform Gradient Clipping. | The Exploding Gradient Problem is the opposite of the Vanishing Gradient Problem. In Deep Neural Networks gradients may explode during backpropagation, resulting number overflows. A common technique to deal with exploding gradients is to perform Gradient Clipping. | ||
• On the difficulty of training recurrent neural networks | • On the difficulty of training recurrent neural networks | ||
Version du 6 mars 2018 à 19:45
Domaine
Vocabulary Apprentissage profond
Définition
Termes privilégiés
Anglais
Exploding Gradient Problem
The Exploding Gradient Problem is the opposite of the Vanishing Gradient Problem. In Deep Neural Networks gradients may explode during backpropagation, resulting number overflows. A common technique to deal with exploding gradients is to perform Gradient Clipping. • On the difficulty of training recurrent neural networks
Contributeurs: Evan Brach, Claude Coulombe, Jacques Barolet, Patrick Drouin, Pierre Labreche, wiki
