RNNbow : Visualizing Learning Via Backpropagation Gradients in RNNs
We present RNNbow, an interactive tool for visualizing the gradient flow during backpropagation in training of recurrent neural networks. By visualizing the gradient, as opposed to activations, RNNbow offers insight into how the network is learning. We show how it illustrates the vanishing gradient...
Description complète
Détails bibliographiques
| Publié dans: | IEEE computer graphics and applications. - 1991. - 38(2018), 6 vom: 01. Nov., Seite 39-50
|
|---|
| Auteur principal: |
Cashman, Dylan
(Auteur) |
|---|
| Autres auteurs: |
Patterson, Genevieve,
Mosca, Abigail,
Watts, Nathan,
Robinson, Shannon,
Chang, Remco |
|---|
| Format: | Article en ligne
|
|---|
| Langue: | English |
|---|
| Publié: |
2018
|
|---|
| Accès à la collection: | IEEE computer graphics and applications
|
|---|
| Sujets: | Journal Article
Research Support, U.S. Gov't, Non-P.H.S. |
|---|