RNNbow: Visualizing Learning Via Backpropagation Gradients in RNNs

被引：22

作者：

Cashman, Dylan ^{[1
]}

Patterson, Genevieve ^{[2
]}

Mosca, Abigail ^{[1
]}

Watts, Nathan ^{[3
]}

Robinson, Shannon ^{[1
]}

Chang, Remco ^{[1
]}

机构：

[1] Tufts Univ, Dept Comp Sci, Medford, MA 02155 USA

[2] Microsoft Res, Bengaluru, India

[3] Tufts Univ, Medford, MA 02155 USA

来源：

IEEE COMPUTER GRAPHICS AND APPLICATIONS | 2018年 / 38卷 / 06期

基金：

美国国家科学基金会;

关键词：

Recurrent neural networks;

D O I：

10.1109/MCG.2018.2878902

中图分类号：

TP31 [计算机软件];

学科分类号：

081202 ; 0835 ;

摘要：

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.

引用

页码：39 / 50

页数：12

共 19 条

[1]

Abadi M, 2016, PROCEEDINGS OF OSDI'16: 12TH USENIX SYMPOSIUM ON OPERATING SYSTEMS DESIGN AND IMPLEMENTATION, P265

[2]

[Anonymous], RECURRENT NEURAL N 3

[3]

[Anonymous], CORR

[4]

[Anonymous], ARXIV13126026 CORR