Recurrent neural networks, and in particular long short-term memory networks
(LSTMs), are a remarkably effective tool for sequence modeling that learn a
dense black-box hidden representation of their sequential input. Researchers
interested in better understanding these models have studied the changes in
hidden state representations over time and noticed some interpretable patterns
but also significant noise. In this work, we present LSTMVis a visual analysis
tool for recurrent neural networks with a focus on understanding these hidden
state dynamics. The tool allows a user to select a hypothesis input range to
focus on local state changes, to match these states changes to similar patterns
in a large data set, and to align these results with domain specific structural
annotations. We further show several use cases of the tool for analyzing
specific hidden state properties on datasets containing nesting, phrase
structure, and chord progressions, and demonstrate how the tool can be used to
isolate patterns for further statistical analysis.
Description
Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks
%0 Generic
%1 strobelt2016visual
%A Strobelt, Hendrik
%A Gehrmann, Sebastian
%A Huber, Bernd
%A Pfister, Hanspeter
%A Rush, Alexander M.
%D 2016
%K LSTM visualization
%T Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks
%U http://arxiv.org/abs/1606.07461
%X Recurrent neural networks, and in particular long short-term memory networks
(LSTMs), are a remarkably effective tool for sequence modeling that learn a
dense black-box hidden representation of their sequential input. Researchers
interested in better understanding these models have studied the changes in
hidden state representations over time and noticed some interpretable patterns
but also significant noise. In this work, we present LSTMVis a visual analysis
tool for recurrent neural networks with a focus on understanding these hidden
state dynamics. The tool allows a user to select a hypothesis input range to
focus on local state changes, to match these states changes to similar patterns
in a large data set, and to align these results with domain specific structural
annotations. We further show several use cases of the tool for analyzing
specific hidden state properties on datasets containing nesting, phrase
structure, and chord progressions, and demonstrate how the tool can be used to
isolate patterns for further statistical analysis.
@misc{strobelt2016visual,
abstract = {Recurrent neural networks, and in particular long short-term memory networks
(LSTMs), are a remarkably effective tool for sequence modeling that learn a
dense black-box hidden representation of their sequential input. Researchers
interested in better understanding these models have studied the changes in
hidden state representations over time and noticed some interpretable patterns
but also significant noise. In this work, we present LSTMVis a visual analysis
tool for recurrent neural networks with a focus on understanding these hidden
state dynamics. The tool allows a user to select a hypothesis input range to
focus on local state changes, to match these states changes to similar patterns
in a large data set, and to align these results with domain specific structural
annotations. We further show several use cases of the tool for analyzing
specific hidden state properties on datasets containing nesting, phrase
structure, and chord progressions, and demonstrate how the tool can be used to
isolate patterns for further statistical analysis.},
added-at = {2016-06-30T00:19:49.000+0200},
author = {Strobelt, Hendrik and Gehrmann, Sebastian and Huber, Bernd and Pfister, Hanspeter and Rush, Alexander M.},
biburl = {https://www.bibsonomy.org/bibtex/217d9da41f5edcbb09ccc91d302b9f37e/tongw},
description = {Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks},
interhash = {ca1dc253457c102b6aadf23f28556b94},
intrahash = {17d9da41f5edcbb09ccc91d302b9f37e},
keywords = {LSTM visualization},
note = {cite arxiv:1606.07461},
timestamp = {2016-06-30T00:19:49.000+0200},
title = {Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks},
url = {http://arxiv.org/abs/1606.07461},
year = 2016
}