-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
309 lines (264 loc) · 12.6 KB
/
index.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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
<!-- Replace the content tag with appropriate information -->
<meta name="description" content="Data Selection | ReDS Lab">
<meta property="og:title" content="Performance Scaling"/>
<meta property="og:description" content="projektor | Performance Scaling via Optimal Transport: Enabling Data Selection from Partially Revealed Sources"/>
<meta property="og:url" content="ruoxi-jia-group.github.io/dataselection"/>
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
<meta property="og:image" content="static/images/projektor1.png" />
<meta property="og:image:width" content="1200"/>
<meta property="og:image:height" content="630"/>
<meta name="twitter:title" content="projektor: Data Selection | ReDS Lab">
<meta name="twitter:description" content="projektor | Performance Scaling via Optimal Transport: Enabling Data Selection from Partially Revealed Sources">
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
<meta name="twitter:image" content="static/images/projektor1.png">
<meta name="twitter:card" content="summary_large_image">
<!-- Keywords for your paper to be indexed by-->
<meta name="keywords" content="data selection; performance scaling; scaling law; optimal transport; data valuation">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>projektor</title>
<link rel="icon" type="image/x-icon" href="static/images/favicon.ico">
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<link rel="stylesheet" href="static/css/bulma.min.css">
<link rel="stylesheet" href="static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="static/css/bulma-slider.min.css">
<link rel="stylesheet" href="static/css/fontawesome.all.min.css">
<link rel="stylesheet"
href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="static/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script src="https://documentcloud.adobe.com/view-sdk/main.js"></script>
<script defer src="static/js/fontawesome.all.min.js"></script>
<script src="static/js/bulma-carousel.min.js"></script>
<script src="static/js/bulma-slider.min.js"></script>
<script src="static/js/index.js"></script>
</head>
<body>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1 publication-title">projektor | Performance Scaling via Optimal Transport: Enabling Data Selection from Partially Revealed Sources</h1>
<div class="is-size-5 publication-authors">
<!-- Paper authors -->
<span class="author-block">
<a href="https://twitter.com/feiyang_ml" target="_blank">Feiyang Kang<sup>1</sup></a><sup>*</sup>,</span>
<span class="author-block">
<a href="https://justhoanganh.com" target="_blank">Hoàng Anh Just<sup>1</sup></a><sup>*</sup>,</span>
<span class="author-block">
<a href="https://anitksahu.github.io/" target="_blank">Anit Kumar Sahu<sup>2</sup></a>,</span>
<span class="author-block">
<a href="https://ruoxijia.info" target="_blank">Ruoxi Jia<sup>1</sup></a></span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>Virginia Tech, <sup>2</sup>Amazon Alexa AI<br>NeurIPS 2023
<br> ICML 2023 Workshop on Data-centric Machine Learning Research</span>
<span class="eql-cntrb"><small><br><sup>*</sup>Indicates Equal Contribution</small></span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- Arxiv PDF link -->
<span class="link-block">
<a href="https://openreview.net/forum?id=quMBEd27x9" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>OpenReview</span>
</a>
</span>
<!-- Github link -->
<span class="link-block">
<a href="https://github.com/ruoxi-jia-group/projektor" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<!-- ArXiv abstract Link -->
<span class="link-block">
<a href="https://arxiv.org/abs/2307.02460" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>ArXiv</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Teaser video-->
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<img src="static/images/web_projektor_abs.png" alt="MY ALT TEXT"/>
<h2 class="subtitle has-text-centered">
Overview of projektor, which take as inputs the public pilot data from each source,
a selection budget, a target model, a validation set representing the test distribution, and return the
optimal combination of data sources as well as the prediction of the resulting model performance.</h2>
</div>
</div>
</section>
<!-- End teaser video -->
<!-- Paper abstract -->
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
<p>
Traditionally, <strong>data selection</strong> has been studied in settings where all samples from
prospective sources are fully revealed to a machine learning developer. However,
in practical data exchange scenarios, data providers often reveal only a <strong>limited
subset of samples</strong> before an acquisition decision is made. Recently, there have
been efforts to fit scaling functions that predict model performance at any size
and data source composition using the limited available samples. However, these
scaling functions are usually black-box, computationally expensive to fit, highly
susceptible to overfitting, or/and difficult to optimize for data selection. This paper
proposes a framework called <strong>projektor</strong>, which <strong>predicts model performance</strong>
and <strong>supports data selection</strong> decisions based on partial samples of prospective data
sources. Our approach distinguishes itself from existing work by introducing a
novel two-stage performance inference process. In the <strong>first stage</strong>, we leverage the
Optimal Transport distance to predict the model’s performance <strong>for any data mixture</strong>
ratio within the range of disclosed data sizes. In the <strong>second stage</strong>, we <strong>extrapolate
the performance</strong> to larger undisclosed data sizes based on a novel <strong>parameter-
free mapping</strong> technique inspired by neural scaling laws. We further derive an
<strong>efficient gradient-based</strong> method to <strong>select data sources</strong> based on the projected model
performance. Evaluation over a diverse range of applications (e.g., vision, text,
fine-tuning, noisy data sources, etc.) demonstrates that projektor significantly
improves existing performance scaling approaches in terms of both the accuracy of
performance inference and the computation costs associated with constructing the
performance predictor. Also, projektor outperforms by a wide margin in data
selection effectiveness compared to a range of other off-the-shelf solutions. We
provide projektor as an <a href="https://github.com/ruoxi-jia-group/projektor" target="_blank"><font color="Green">open-source toolkit</font></a>.</p>
</div>
</div>
</div>
</div>
</section>
<!-- End paper abstract -->
<!-- Image carousel -->
<section class="hero is-small">
<div class="hero-body">
<div class="container">
<h2 class="title is-3">projektor | Data Selection Pipeline</h2>
<div id="results-carousel" class="carousel results-carousel">
<div class="item">
<!-- Your image here -->
<img src="static/images/projektor1.png" alt="MY ALT TEXT"/>
<h2 class="subtitle has-text-centered">
Problem setup of <strong>projektor</strong>. 1/5
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/projektor2.png" alt="MY ALT TEXT"/>
<h2 class="subtitle has-text-centered">
Performance Predictor For a Fixed Small Scale. 2/5
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/projektor3.png" alt="MY ALT TEXT"/>
<h2 class="subtitle has-text-centered">
Performance Projection to Larger Scale N. 3/5
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/projektor4.png" alt="MY ALT TEXT"/>
<h2 class="subtitle has-text-centered">
Efficient Data Composition Optimization. 4/5
.
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/projektor5.png" alt="MY ALT TEXT"/>
<h2 class="subtitle has-text-centered">
Data Selection with <strong>projektor</strong>. 5/5
</h2>
</div>
</div>
</div>
</div>
</section>
<!-- End image carousel -->
<!-- Youtube video -->
<!-- <section class="hero is-small is-light">
<div class="hero-body">
<div class="container"> -->
<!-- Paper video. -->
<!-- <h2 class="title is-3">Video Presentation</h2>
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="publication-video"> -->
<!-- Youtube embed code here -->
<!-- <iframe src="https://www.youtube.com/embed/JkaxUblCGz0" frameborder="0" allow="autoplay; encrypted-media" allowfullscreen></iframe>
</div>
</div>
</div>
</div>
</div>
</section> -->
<!-- End youtube video -->
<!-- Paper poster -->
<section class="hero is-small is-light">
<div class="hero-body">
<div class="container">
<h2 class="title">Project Poster (NeurIPS 2023)</h2>
<iframe src="static/pdfs/NeurIPS'23_projektor.pdf" width="100%" height="550">
</iframe>
</div>
</div>
</section>
<!--End paper poster -->
<!--BibTex citation -->
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>
@inproceedings{
kang2023performance,
title={Performance Scaling via Optimal Transport: Enabling Data Selection from Partially Revealed Sources},
author={Feiyang Kang and Hoang Anh Just and Anit Kumar Sahu and Ruoxi Jia},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=quMBEd27x9}
}
</code></pre>
</div>
</section>
<!--End BibTex citation -->
<footer class="footer">
<div class="container">
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p>
<a href="https://ruoxijia.info/reds" target="_blank">ReDS Lab @ 2023</a> | We greatly appreciate the author for providing the <a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">template</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
<!-- Statcounter tracking code -->
<!-- You can add a tracker to track page visits by creating an account at statcounter.com -->
<!-- End of Statcounter Code -->
</body>
</html>