This repository has been archived by the owner on Jul 27, 2022. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 19
/
Copy pathinterpret.html
139 lines (119 loc) · 7.17 KB
/
interpret.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
---
---
<!DOCTYPE html>
<html lang="en-us">
<head>
{% include meta.html %}
<title>AllenNLP Interpret</title>
<!-- Bootstrap core CSS -->
<link rel="stylesheet" href="https://stackpath.bootstrapcdn.com/bootstrap/4.1.2/css/bootstrap.min.css"
integrity="sha384-Smlep5jCw/wG7hdkwQ/Z5nLIefveQRIY9nfy6xoR1uRYBtpZgI6339F5dgvm/e9B" crossorigin="anonymous">
<style>
p,
li,
th,
td {
font-size: 1.5em;
}
.constrained--med {
max-width: 77em;
padding: 1.25em 1.125em 3.75em 1.125em;
}
.headshot {
width: 12em;
margin: 0.2rem;
}
</style>
</head>
<body id="top">
<div id="page-content">
{% include header.html %}
<div class="banner banner--interior-hero">
<div class="constrained constrained--sm">
<div class="banner--interior-hero__content">
<h2>AllenNLP Interpret:<br> A Framework for Explaining Predictions of NLP Models</h2>
<h5>Eric Wallace, Jens Tuyls, Junlin Wang, Sanjay Subramanian, <br> Matt Gardner, and Sameer Singh</h5><br>EMNLP 2019 Demo
</div>
</div>
</div>
<div class="constrained constrained--med">
<p>
Despite constant advances and seemingly super-human performance on constrained domains, state-of-the-art models for NLP are imperfect. These imperfections, coupled with today's advances being driven by (seemingly black-box) neural models, leave researchers and practitioners scratching their heads asking, <i>why did my model make this prediction?</i>
</p>
<p>
We present AllenNLP Interpret, a toolkit built on top of AllenNLP for interactive model interpretations. The toolkit makes it easy to apply gradient-based <b><u>saliency maps</b></u> and <u><b>adversarial attacks</b></u> to <i>new models</i>, as well as develop <i>new interpretation methods</i>. AllenNLP interpret contains three components: a suite of interpretation techniques applicable to most models, APIs for developing new interpretation methods (e.g., APIs to obtain input gradients), and reusable front-end components for visualizing the interpretation results.
</p>
<p>This page presents links to:</p>
<li><a href="https://arxiv.org/abs/1909.09251", target="_blank">Paper</a> describing the framework, the technical implementation details, and showing some example use cases.</li>
<li>Live demos for various models and tasks, such as
<ul>
<li><a href="https://demo.allennlp.org/masked-lm", target="_blank">Masked Language Modeling</a> using BERT, to explain why it made certain mask predictions.</li>
<div style="margin:auto"><img src="assets/interpret-photos/bert-combined.jpg" style="width:65em"></div>
<li><a href="https://demo.allennlp.org/textual-entailment/MTA1Mjc2Ng==", target="_blank">Textual Entailment</a> and <a href="https://demo.allennlp.org/sentiment-analysis/MTA1Mjk1Mg==", target="_blank">Sentiment Analysis</a> using ELMo-based LSTM classifiers.</li>
<div style="text-align:center"><img src="assets/interpret-photos/reduction.png" style="width:65em"></div>
<li>SQuAD and DROP <a href="https://demo.allennlp.org/reading-comprehension/MTA1Mjc3NA==", target="_blank">reading comprehension</a> using an ELMo-based QANet</li>
<div style="text-align:center"><img src="assets/interpret-photos/squad_screenshot.png" style="width:65em"></div>
<li><a href="https://demo.allennlp.org/named-entity-recognition/MTA2MTUyNg==", target="_blank">NER</a> using an LSTM-CRF model based on ELMo.</li>
<div style="text-align:center"><img src="assets/interpret-photos/ner.png" style="width:40em"></div>
</ul>
</li>
<li>Tutorials for interpreting any <a href="https://github.com/allenai/allennlp-demo#contributing-a-new-allennlp-model-to-the-demo", target="_blank">model of your choice</a>, and addding <a href="https://github.com/allenai/allennlp-demo#adding-a-new-interpretation-method", target="_blank">a new interpretation method</a>.</li>
<li><a href="https://github.com/allenai/allennlp/tree/master/allennlp/interpret", target="_blank">Code</a> for interpreting/attacking models and visualizing the results in the demo (e.g., <a href="https://github.com/allenai/allennlp-demo/blob/master/demo/src/components/demos/SentimentAnalysis.js", target="_blank">sentiment analysis</a>).</li>
<p><i>Citation:</i>
<pre> @inproceedings{Wallace2019AllenNLP,
Author = {Eric Wallace and Jens Tuyls and Junlin Wang and Sanjay Subramanian
and Matt Gardner and Sameer Singh},
Booktitle = {Empirical Methods in Natural Language Processing},
Year = {2019},
Title = { {AllenNLP Interpret}: A Framework for Explaining Predictions of {NLP} Models}}
</pre>
</p>
<br>
<div id="authors" class="anchor-target">
<h3>Paper Authors</h3>
<div class="row" style="margin: 0 auto;">
<div class="card headshot box-shadow">
<img class="card-img-top" src="assets/interpret-photos/eric.jpg" alt="eric">
<div class="card-body">
<p class="card-text"><a href="http://www.ericswallace.com">Eric Wallace</a></p>
</div>
</div>
<div class="card headshot box-shadow">
<img class="card-img-top" src="assets/interpret-photos/jens.jpg" alt="jens">
<div class="card-body">
<p class="card-text">Jens Tuyls</a></p>
</div>
</div>
<div class="card headshot box-shadow">
<img class="card-img-top" src="assets/interpret-photos/junlin.png" alt="junlin">
<div class="card-body">
<p class="card-text">Junlin Wang</p>
</div>
</div>
<div class="card headshot box-shadow">
<img class="card-img-top" src="assets/interpret-photos/sanjay.png" alt="matt">
<div class="card-body">
<p class="card-text"><a href="https://sanjayss34.github.io/" style="font-size: 1em">Sanjay S</a></p>
</div>
</div>
<div class="card headshot box-shadow">
<img class="card-img-top" src="assets/interpret-photos/matt.jpg" alt="matt">
<div class="card-body">
<p class="card-text"><a href="https://matt-gardner.github.io/">Matt Gardner</a></p>
</div>
</div>
<div class="card headshot box-shadow">
<img class="card-img-top" src="assets/interpret-photos/sameer.jpg" alt="sameer">
<div class="card-body">
<p class="card-text"><a href="http://sameersingh.org/">Sameer Singh</a></p>
</div>
</div>
</div>
</div>
</div>
{% include footer.html %}
</div>
{% include svg-sprite.html %}
{% include scripts.html %}
</body>
</html>