-
Notifications
You must be signed in to change notification settings - Fork 0
/
paper_4.html
426 lines (305 loc) · 19 KB
/
paper_4.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
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
<!DOCTYPE html>
<html lang="en">
<head>
<!-- Required meta tags -->
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no" />
<!-- External Javascript libs_ext -->
<script src="https://cdn.jsdelivr.net/npm/d3@5/dist/d3.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/handlebars@4.7.3/dist/handlebars.min.js"
integrity="sha256-/PJBs6QWvXijOFIX04kZpLb6ZtSQckdOIavLWKKOgXU=" crossorigin="anonymous"></script>
<script src="https://code.jquery.com/jquery-3.6.0.min.js"
integrity="sha256-/xUj+3OJU5yExlq6GSYGSHk7tPXikynS7ogEvDej/m4=" crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/popper.js@1.16.0/dist/umd/popper.min.js"
integrity="sha384-Q6E9RHvbIyZFJoft+2mJbHaEWldlvI9IOYy5n3zV9zzTtmI3UksdQRVvoxMfooAo"
crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/bootstrap@4.4.1/dist/js/bootstrap.min.js"
integrity="sha256-WqU1JavFxSAMcLP2WIOI+GB2zWmShMI82mTpLDcqFUg=" crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/moment@2.24.0/min/moment.min.js"
integrity="sha256-4iQZ6BVL4qNKlQ27TExEhBN1HFPvAvAMbFavKKosSWQ=" crossorigin="anonymous"></script>
<script src="https://cdn.jsdelivr.net/npm/moment-timezone@0.5.28/builds/moment-timezone-with-data.min.js"
integrity="sha256-IWYg4uIC8/erItNXYvLtyYHioRi2zT1TFva8qaAU/ww=" crossorigin="anonymous"></script>
<!-- Library libs_ext -->
<script src="static/js/libs_ext/typeahead.bundle.js"></script>
<script src="static/js/tagEditor/jquery.caret.min.js"></script>
<script src="static/js/tagEditor/jquery.tag-editor.min.js"></script>
<!-- Internal Libs -->
<script src="static/js/data/api.js"></script>
<!-- External CSS -->
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/bootstrap@4.3.1/dist/css/bootstrap.min.css"
integrity="sha256-YLGeXaapI0/5IgZopewRJcFXomhRMlYYjugPLSyNjTY=" crossorigin="anonymous">
<!-- External Fonts (no google for china) -->
<link href="static/css/Lato.css" rel="stylesheet" />
<link href="static/css/Exo.css" rel="stylesheet" />
<link href="static/css/Cuprum.css" rel="stylesheet" />
<link rel="stylesheet" href="static/css/main.css" />
<!-- <link rel="stylesheet" href="static/css/fa_regular.css"/>-->
<link rel="stylesheet" href="static/css/fa_solid.css" />
<link rel="stylesheet" href="static/css/lazy_load.css" />
<link rel="stylesheet" href="static/css/typeahead.css" />
<script type="text/javascript" src="https://cdn.jsdelivr.net/npm/daterangepicker/daterangepicker.min.js"></script>
<link rel="stylesheet" type="text/css" href="https://cdn.jsdelivr.net/npm/daterangepicker/daterangepicker.css" />
<title>Machine Learning Simulation: Predicting Physics in Mesh-reduced Space with Temporal Attention</title>
<link rel="icon" type="image/png" href="favicon.ico">
<!-- Google Analytics -->
<!-- <script async src="https://www.googletagmanager.com/gtag/js?id=G-GT7LEVW06Z"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() { dataLayer.push(arguments); }
gtag('js', new Date());
gtag('config', 'G-GT7LEVW06Z');
</script> -->
<!-- head block ends here-->
<meta name="citation_title" content="Predicting Physics in Mesh-reduced Space with Temporal Attention" />
<meta name="citation_author" content="Xu Han" />
<meta name="citation_author" content="Han Gao" />
<meta name="citation_author" content="Tobias Pfaff" />
<meta name="citation_author" content="Jian-Xun Wang" />
<meta name="citation_author" content="Li-Ping Liu" />
<meta name="citation_abstract" content="Graph-based next-step prediction models have recently been very successful in modeling complex high-dimensional physical systems on irregular meshes. However, due to their short temporal attention span, these models suffer from error accumulation and drift. In this paper, we propose a new method that captures long-term dependencies through a transformer-style temporal attention model. We introduce an encoder-decoder structure to summarize features and create a compact mesh representation of the system state, to allow the temporal model to operate on a low-dimensional mesh representations in a memory efficient manner. Our method outperforms a competitive GNN baseline on several complex fluid dynamics prediction tasks, from sonic shocks to vascular flow. We demonstrate stable rollouts without the need for training noise and show perfectly phase-stable predictions even for very long sequences. More broadly, we believe our approach paves the way to bringing the benefits of attention-based sequence models to solving high-dimensional complex physics tasks." />
<meta name="citation_keywords" content="Graph Neural Networks" />
<meta name="citation_keywords" content="transformers" />
<meta name="citation_keywords" content="autoregressive" />
<meta name="citation_keywords" content="multi-head attention" />
<meta name="citation_keywords" content="temporal attention" />
<meta name="citation_keywords" content="GNNs" />
<meta name="citation_keywords" content="message passing" />
<meta name="citation_keywords" content="encoder-processor-decoder" />
<meta name="citation_keywords" content="computational efficiency" />
<meta name="citation_keywords" content="neural network simulators" />
<meta name="citation_keywords" content="computational fluid dynamics" />
<meta name="citation_keywords" content="CFD simulation" />
<meta name="citation_keywords" content="cylinderflow" />
<meta name="citation_keywords" content="mesh" />
<meta name="citation_keywords" content="graph mesh reducer" />
<meta name="citation_keywords" content="autoencoder" />
<meta name="citation_pdf_url" content="https://arxiv.org/pdf/2201.09113.pdf" />
<script type="text/javascript" src="static/js/vis/vis-network.min.js"></script>
<style>
.red-hyper-link {
color: #ED1C24 !important;
}
</style>
</head>
<body>
<!-- NAV -->
<!-- add the below items to the "navigation_bar" list if those pages are needed; can't comment those out b/c jinja doesn't support comment
('index.html', 'Home'),
('papers.html', 'Advanced Search'),
('add_paper.html', 'Add Paper'),
('join_us.html', 'Join Us'),
('faq.html', 'FAQ')
-->
<nav class="navbar sticky-top navbar-expand-lg navbar-light" style="background-color: #EFECE5;" id="main-nav">
<div class="container">
<a class="navbar-brand" href="index.html">
<img class="logo" src="static/images/mls_logo_withfont.svg" style="width: 250px;"/>
</a>
<button class="navbar-toggler" type="button" data-toggle="collapse" data-target="#navbarNav"
aria-controls="navbarNav" aria-expanded="false" aria-label="Toggle navigation">
<span class="navbar-toggler-icon"></span>
</button>
<div class="collapse navbar-collapse text-right flex-grow-1" id="navbarNav">
<ul class="navbar-nav ml-auto">
<li class="nav-item ">
<a class="nav-link" href="index.html" style="font-family: 'Ubuntu', sans-serif;">Home</a>
</li>
<li class="nav-item ">
<a class="nav-link" href="add_paper.html" style="font-family: 'Ubuntu', sans-serif;">Add Paper</a>
</li>
</ul>
</div>
</div>
</nav>
<!-- User Overrides -->
<!-- top block ends here-->
<div class="container">
<!-- Tabs -->
<div class="tabs">
</div>
<!-- Content -->
<div class="content">
<!-- Title -->
<div class="pp-card m-3">
<div class="card-header">
<h2 class="card-title main-title text-center" style="color: black;">
Predicting Physics in Mesh-reduced Space with Temporal Attention
</h2>
<h3 class="card-subtitle mb-2 text-muted text-center">
<a href="papers.html?author=Xu Han" target="_blank"
data-tippy-content="See all papers authored by Xu Han"
class="text-muted filterByAuthorLink">Xu Han</a>,
<a href="papers.html?author=Han Gao" target="_blank"
data-tippy-content="See all papers authored by Han Gao"
class="text-muted filterByAuthorLink">Han Gao</a>,
<a href="papers.html?author=Tobias Pfaff" target="_blank"
data-tippy-content="See all papers authored by Tobias Pfaff"
class="text-muted filterByAuthorLink">Tobias Pfaff</a>,
<a href="papers.html?author=Jian-Xun Wang" target="_blank"
data-tippy-content="See all papers authored by Jian-Xun Wang"
class="text-muted filterByAuthorLink">Jian-Xun Wang</a>,
<a href="papers.html?author=Li-Ping Liu" target="_blank"
data-tippy-content="See all papers authored by Li-Ping Liu"
class="text-muted filterByAuthorLink">Li-Ping Liu</a>
</h3>
<h3 class="card-subtitle mb-2 text-muted text-center">
26/5/2022
</h3>
<div id="citation-code-wrapper" class="text-muted text-center"
>
</div>
<p class="text-center" style="margin-bottom: 0px;">
<span>Keywords:</span>
<a href="papers.html?keyword=Graph Neural Networks" target="_blank"
data-tippy-content="See all papers with keyword 'Graph Neural Networks'"
class="text-secondary text-decoration-none filterByKeywordLink">Graph Neural Networks</a>,
<a href="papers.html?keyword=transformers" target="_blank"
data-tippy-content="See all papers with keyword 'transformers'"
class="text-secondary text-decoration-none filterByKeywordLink">transformers</a>,
<a href="papers.html?keyword=autoregressive" target="_blank"
data-tippy-content="See all papers with keyword 'autoregressive'"
class="text-secondary text-decoration-none filterByKeywordLink">autoregressive</a>,
<a href="papers.html?keyword=multi-head attention" target="_blank"
data-tippy-content="See all papers with keyword 'multi-head attention'"
class="text-secondary text-decoration-none filterByKeywordLink">multi-head attention</a>,
<a href="papers.html?keyword=temporal attention" target="_blank"
data-tippy-content="See all papers with keyword 'temporal attention'"
class="text-secondary text-decoration-none filterByKeywordLink">temporal attention</a>,
<a href="papers.html?keyword=GNNs" target="_blank"
data-tippy-content="See all papers with keyword 'GNNs'"
class="text-secondary text-decoration-none filterByKeywordLink">GNNs</a>,
<a href="papers.html?keyword=message passing" target="_blank"
data-tippy-content="See all papers with keyword 'message passing'"
class="text-secondary text-decoration-none filterByKeywordLink">message passing</a>,
<a href="papers.html?keyword=encoder-processor-decoder" target="_blank"
data-tippy-content="See all papers with keyword 'encoder-processor-decoder'"
class="text-secondary text-decoration-none filterByKeywordLink">encoder-processor-decoder</a>,
<a href="papers.html?keyword=computational efficiency" target="_blank"
data-tippy-content="See all papers with keyword 'computational efficiency'"
class="text-secondary text-decoration-none filterByKeywordLink">computational efficiency</a>,
<a href="papers.html?keyword=neural network simulators" target="_blank"
data-tippy-content="See all papers with keyword 'neural network simulators'"
class="text-secondary text-decoration-none filterByKeywordLink">neural network simulators</a>,
<a href="papers.html?keyword=computational fluid dynamics" target="_blank"
data-tippy-content="See all papers with keyword 'computational fluid dynamics'"
class="text-secondary text-decoration-none filterByKeywordLink">computational fluid dynamics</a>,
<a href="papers.html?keyword=CFD simulation" target="_blank"
data-tippy-content="See all papers with keyword 'CFD simulation'"
class="text-secondary text-decoration-none filterByKeywordLink">CFD simulation</a>,
<a href="papers.html?keyword=cylinderflow" target="_blank"
data-tippy-content="See all papers with keyword 'cylinderflow'"
class="text-secondary text-decoration-none filterByKeywordLink">cylinderflow</a>,
<a href="papers.html?keyword=mesh" target="_blank"
data-tippy-content="See all papers with keyword 'mesh'"
class="text-secondary text-decoration-none filterByKeywordLink">mesh</a>,
<a href="papers.html?keyword=graph mesh reducer" target="_blank"
data-tippy-content="See all papers with keyword 'graph mesh reducer'"
class="text-secondary text-decoration-none filterByKeywordLink">graph mesh reducer</a>,
<a href="papers.html?keyword=autoencoder" target="_blank"
data-tippy-content="See all papers with keyword 'autoencoder'"
class="text-secondary text-decoration-none filterByKeywordLink">autoencoder</a>
</p>
<p class="text-center" style="margin-bottom: 0px;">
<span>Venue: </span>
<a href="papers.html?venue=ICLR" target="_blank" class="text-secondary text-decoration-none">ICLR 2022</a>
</p>
<div class="text-center p-3">
<a class="card-link red-hyper-link" target="_blank" href="https://arxiv.org/pdf/2201.09113.pdf">
Paper
</a>
<a class="card-link red-hyper-link" data-toggle="collapse" role="button" href="#citation">
Citation
</a>
</div>
</div>
<span id="invisible-paper-id" style="display: none;">4</span>
</div>
<div id="citation" class="pp-card m-3 collapse">
<div class="card-body">
<div class="card-text">
<span style="font-size: large; font-weight: bold;">Bibtex:</span>
<span style="white-space: pre-line; position: relative; left: 20px;">
@article{DBLP:journals/corr/abs-2201-09113,
author = {Xu Han and
Han Gao and
Tobias Pffaf and
Jian{-}Xun Wang and
Li{-}Ping Liu},
title = {Predicting Physics in Mesh-reduced Space with Temporal Attention},
journal = {CoRR},
volume = {abs/2201.09113},
year = {2022},
url = {https://arxiv.org/abs/2201.09113},
eprinttype = {arXiv},
eprint = {2201.09113},
timestamp = {Tue, 01 Feb 2022 14:59:01 +0100},
biburl = {https://dblp.org/rec/journals/corr/abs-2201-09113.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
</span>
</div>
<p></p>
</div>
</div>
<div id="details" class="pp-card m-3">
<div class="card-body">
<div class="card-text">
<div id="abstractExample">
<p style="font-weight: bolder; font-size: 25px; text-align: center;">Abstract</p>
Graph-based next-step prediction models have recently been very successful in modeling complex high-dimensional physical systems on irregular meshes. However, due to their short temporal attention span, these models suffer from error accumulation and drift. In this paper, we propose a new method that captures long-term dependencies through a transformer-style temporal attention model. We introduce an encoder-decoder structure to summarize features and create a compact mesh representation of the system state, to allow the temporal model to operate on a low-dimensional mesh representations in a memory efficient manner. Our method outperforms a competitive GNN baseline on several complex fluid dynamics prediction tasks, from sonic shocks to vascular flow. We demonstrate stable rollouts without the need for training noise and show perfectly phase-stable predictions even for very long sequences. More broadly, we believe our approach paves the way to bringing the benefits of attention-based sequence models to solving high-dimensional complex physics tasks.
</div>
</div>
<p></p>
</div>
</div>
<!-- Youtube Video -->
<!-- Project Webpage -->
<!-- Citation Graph -->
<!-- <div class="border-top my-3"></div>
<div class="row p-4">
<div class="col-12" style="text-align: center;">
<span style="font-size: 25px;">Citation Graph</span> <br>
<span>(Double click on nodes to open corresponding papers' pages)</span>
</div>
</div>
<div id="citationGraph" style="width: 100%;
height: 600px;
border: 1px solid lightgray;" ondblclick="openPaperLink()"></div>
<p><span style="font-size: 10px;">*</span> Showing citation graph for papers within our database. Data retrieved from <a href="https://www.semanticscholar.org/search?q=Predicting Physics in Mesh-reduced Space with Temporal Attention&sort=relevance">Semantic Scholar</a>. For full citation graphs, visit <a href="https://www.connectedpapers.com/search?q=Predicting Physics in Mesh-reduced Space with Temporal Attention">ConnectedPapers</a>.</p>
<script src="https://cdn.jsdelivr.net/npm/@popperjs/core@2.4.0/dist/umd/popper.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/tippy.js@6/dist/tippy-bundle.umd.min.js"></script>
<script src="static/js/views/paper_detail.js"></script>
<script>
tippy(".filterByAuthorLink");
tippy(".filterByKeywordLink");
const paperID = document.getElementById("invisible-paper-id").innerHTML;
drawCitationGraphAndGenerateCitationCode(paperID);
</script>
-->
</div>
</div>
<!-- body block ends here-->
<!-- Footer -->
<footer class="footer"
style="background-color: #353535; width: 100%; height: 4vh; display: flex; justify-content: center; align-items: center; ">
<!-- no contents needed -->
</footer>
<!-- Code for hash tags -->
<script type="text/javascript">
$(document).ready(function () {
if (window.location.hash !== "") {
$(`a[href="${window.location.hash}"]`).tab("show");
}
$("a[data-toggle='tab']").on("shown.bs.tab", function (e) {
const hash = $(e.target).attr("href");
if (hash.substr(0, 1) === "#") {
const position = $(window).scrollTop();
window.location.replace(`#${hash.substr(1)}`);
$(window).scrollTop(position);
}
});
});
</script>
<!-- footer block ends here -->
</body>
</html>