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...
| Veröffentlicht in: | IEEE computer graphics and applications. - 1991. - 38(2018), 6 vom: 01. Nov., Seite 39-50 |
|---|---|
| 1. Verfasser: | |
| Weitere Verfasser: | , , , , |
| Format: | Online-Aufsatz |
| Sprache: | English |
| Veröffentlicht: |
2018
|
| Zugriff auf das übergeordnete Werk: | IEEE computer graphics and applications |
| Schlagworte: | Journal Article Research Support, U.S. Gov't, Non-P.H.S. |
| Zusammenfassung: | 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 and the training process |
|---|---|
| Beschreibung: | Date Completed 19.03.2019 Date Revised 19.03.2019 published: Print Citation Status PubMed-not-MEDLINE |
| ISSN: | 1558-1756 |
| DOI: | 10.1109/MCG.2018.2878902 |