forked from jonbarron/website
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
356 lines (333 loc) · 14.8 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
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
<!DOCTYPE HTML>
<html lang="en">
<head>
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-MMEEFPHQW8"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-MMEEFPHQW8');
</script>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
<title>Yihao Quan</title>
<meta name="author" content="Yihao Quan">
<meta name="viewport" content="width=device-width, initial-scale=1">
<link rel="shortcut icon" href="images/favicon/favicon.ico" type="image/x-icon">
<link rel="stylesheet" type="text/css" href="stylesheet.css">
</head>
<body>
<table style="width:100%;max-width:800px;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody>
<tr style="padding:0px">
<td style="padding:0px">
<table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody>
<tr style="padding:0px">
<td style="padding:2.5%;width:63%;vertical-align:middle">
<p class="name" style="text-align: center;">
Yihao Quan
</p>
<p>I'm a fourth-year ungraduate student at <a href="https://www.rit.edu/">Rochester Institute of Technology</a>, where I major in Management Information system.
</p>
<p>
Recently, I'm a research intern at <a href="https://pradalab1.github.io/index.html">King Abdullah University of Science and Technology (KAUST) PRADA Lab</a>.
</p>
<p style="text-align:center">
<a href="mailto:[email protected]">Email</a> /
<!-- <a href="data/JonBarron-CV.pdf">CV</a> /
<a href="data/JonBarron-bio.txt">Bio</a> / -->
<a href="https://scholar.google.com/citations?view_op=list_works&hl=en&hl=en&user=Rm69hkYAAAAJ">Scholar</a> /
<!-- <a href="https://www.threads.net/@jonbarron">Threads</a> /
<a href="https://bsky.app/profile/jonbarron.bsky.social">Bluesky</a> / -->
<!-- <a href="https://twitter.com/jon_barron">Twitter</a> / -->
<a href="https://www.linkedin.com/in/yihao-quan-24223632a/">Linkedin</a> /
<a href="https://github.com/itsqyh">Github</a>
</p>
</td>
<td style="padding:2.5%;width:40%;max-width:40%">
<a href="images/img1.jpg"><img style="width:100%;max-width:100%;object-fit: cover; border-radius: 50%;" alt="profile photo" src="images/img1.jpg" class="hoverZoomLink"></a>
</td>
</tr>
</tbody></table>
<table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody>
<tr>
<td style="padding:20px;width:100%;vertical-align:middle">
<h2>Research Interests</h2>
<p> My research primarily focuses on <strong>Trustworthy & Interpretability MLLMs/LLMs</strong>, with a emphasis on the following areas: </p>
<ul> <li><strong>Vision + Language:</strong>
Bridging the gap between our knowledge of language and vision representation, providing a foundation for more interpretable and controllable multimodal systems. </li>
<p>
</p>
<li><strong>Mechanistic Interpretability:</strong>
Understanding the internal mechanisms of LLMs/MLLMs (e.g. circuit analysis, causal tracing, neuron analysis).</li>
<p>
</p>
<li><strong>Hallucination, Factuality and Safety:</strong>
Using the interpretability findings to help downstream tasks (e.g. factual knowledge, enhancing reasoning, reducing hallucinations, model editing), and design safer models.</l>
<p>Representative papers are <span class="highlight">highlighted</span>, * = Equal Contribution</p>
</ul>
</td>
</tr>
<!-- News Section -->
<tr>
<td style="padding:20px;width:100%;vertical-align:middle">
<h2>News</h2>
<ul>
<li><strong>Jan 23th 2025:</strong> Our paper "<a href="https://arxiv.org/abs/2406.06579">From Redundancy to Relevance: Information Flow in LVLMs Across Reasoning Tasks</a>" got accepted by NAACL 2025 Main track.</li>
<li><strong>Dec 13th 2024:</strong> Our paper "<a href="https://arxiv.org/abs/2406.06579">From Redundancy to Relevance: Information Flow in LVLMs Across Reasoning Tasks</a>" received meta-score 4 and was suggested for NAACL'25 main track.</li>
<li><strong>Dec 10th 2024:</strong> Our paper "<a href="https://arxiv.org/abs/2412.09817">Enhancing LVLMs’ Complex Reasoning via Similarity Computation</a>" got accepted by AAAI 2025.</li>
<li><strong>Nov 15th 2024:</strong> Submitted our work "<a href="https://arxiv.org/abs/2411.09968">Seeing Clearly by Layer Two: Enhancing Attention Heads to Alleviate Hallucination in LVLMs</a>" to CVPR'25.</li>
</ul>
</td>
</tr>
</tbody></table>
<table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody>
<tr onmouseout="eah_stop()" onmouseover="eah_start()">
<td style="padding:20px;width:25%;vertical-align:middle; text-align:center">
<div class="one" style="display: inline-block;">
<div class="two" id='eah_image' style="opacity: 0; transition: opacity 0.3s;"><video width=100% muted autoplay loop>
<source src="images/fig1.png" type="video/mp4">
Your browser does not support the video tag.
</video></div>
<img src='images/fig1.png' width=100%>
</div>
<script type="text/javascript">
function eah_start() {
document.getElementById('eah_image').style.opacity = "1";
}
function eah_stop() {
document.getElementById('eah_image').style.opacity = "0";
}
eah_stop()
</script>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://arxiv.org/abs/2411.09968">
<span class="papertitle">Seeing Clearly by Layer Two: Enhancing Attention Heads to Alleviate Hallucination in LVLMs</span>
</a>
<br>
Xiaofeng Zhang*,
<strong><u>Yihao Quan*</u></strong>,
Chaochen Gu,
Chen Shen,
Xiaosong Yuan,
Shaotian Yan,
Jieping Ye
<br>
<a href="https://arxiv.org/abs/2411.09968">arXiv</a>
<br>
<em>CVPR'25 Under Review
<br>
<p></p>
<p>
A plug-and-play and training-free method showing significant hallucination-mitigating performance on different VLMs and metrics.
</p>
</td>
</tr>
<tr onmouseout="information_flow_stop()" onmouseover="information_flow_start()">
<td style="padding:20px;width:25%;vertical-align:middle">
<div class="one">
<div class="two" id='information-flow'>
<video width=100% muted autoplay loop>
<source src="images/llava-cam.png" type="video/mp4">
Your browser does not support the video tag.
</video></div>
<img src='images/llava-cam.png' width=100%>
</div>
<script type="text/javascript">
function information_flow_start() {
document.getElementById('information-flow').style.opacity = "1";
}
function information_flow_stop() {
document.getElementById('information-flow').style.opacity = "0";
}
information_flow_stop()
</script>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://arxiv.org/abs/2406.06579">
<span class="papertitle">From Redundancy to Relevance: Information Flow in LVLMs Across Reasoning Tasks</span>
</a>
<br>
Xiaofeng Zhang*,
<strong><u>Yihao Quan*</u></strong>,
Chaochen Gu,
Chen Shen,
Xiaosong Yuan,
Shaotian Yan,
Liang Xie,
Wenxiao Wang,
Hao Tang,
Jieping Ye
<br>
<a href="https://arxiv.org/abs/2406.06579">arXiv</a>
<br>
<em>Accepted by NAACL'25 Main
<br>
<p></p>
<p>
A novel perspective to enhance understanding of LVLMs and their functioning, particularly for complex reasoning tasks.
</p>
</td>
</tr>
<tr onmouseout="Simignore_stop()" onmouseover="Simignore_start()">
<td style="padding:20px;width:25%;vertical-align:middle">
<div class="one">
<div class="two" id='Simignore_image'><video width=100% muted autoplay loop>
<source src="images/structure-v1.png" type="video/mp4">
Your browser does not support the video tag.
</video></div>
<img src='images/structure-v1.png' width=100%>
</div>
<script type="text/javascript">
function Simignore_start() {
document.getElementById('Simignore_image').style.opacity = "1";
}
function Simignore_stop() {
document.getElementById('Simignore_image').style.opacity = "0";
}
Simignore_stop()
</script>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://github.com/FanshuoZeng/Simignore">
<span class="papertitle">Enhancing LVLMs’ Complex Reasoning via Similarity Computation</span>
</a>
<br>
Fanshuo Zeng*,
Xiaofeng Zhang*,
<strong><u>Yihao Quan</u></strong>,
Zheng Hui,
Jiawei Yao
<br>
<a href="https://github.com/FanshuoZeng/Simignore">Github</a> / <a href="https://arxiv.org/abs/2412.09817">arXiv</a>
<br>
<em> Accepted by AAAI'25
<br>
<p></p>
<p>
A novel image token reduction method, Simignore, designed to enhance the complex reasoning capabilities.
</p>
</td>
</tr>
<!-- <tr onmouseout="recon_stop()" onmouseover="recon_start()" bgcolor="#ffffd0">
<td style="padding:20px;width:25%;vertical-align:middle">
<div class="one">
<div class="two" id='recon_image'><video width=100% height=100% muted autoplay loop>
<source src="images/recon.mp4" type="video/mp4">
Your browser does not support the video tag.
</video></div>
<img src='images/recon.png' width="160">
</div>
<script type="text/javascript">
function recon_start() {
document.getElementById('recon_image').style.opacity = "1";
}
function recon_stop() {
document.getElementById('recon_image').style.opacity = "0";
}
recon_stop()
</script>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://reconfusion.github.io/">
<span class="papertitle">ReconFusion: 3D Reconstruction with Diffusion Priors</span>
</a>
<br>
<a href="https://www.cs.columbia.edu/~rundi/">Rundi Wu*</a>,
<a href="https://bmild.github.io/">Ben Mildenhall*</a>,
<a href="https://henzler.github.io/">Philipp Henzler</a>,
<a href="https://keunhong.com/">Keunhong Park</a>,
<a href="https://ruiqigao.github.io/">Ruiqi Gao</a>,
<a href="https://scholar.google.com/citations?user=_pKKv2QAAAAJ&hl=en/">Daniel Watson</a>,
<a href="https://pratulsrinivasan.github.io/">Pratul P. Srinivasan</a>,
<a href="https://dorverbin.github.io/">Dor Verbin</a>,
<strong>Jonathan T. Barron</strong>,
<a href="https://poolio.github.io/">Ben Poole</a>,
<a href="https://holynski.org/">Aleksander Holynski*</a>
<br>
<em>arXiv</em>, 2023
<br>
<a href="https://reconfusion.github.io/">project page</a>
/
<a href="https://arxiv.org/abs/">arXiv</a>
<p></p>
<p>
Using a multi-image diffusion model as a regularizer lets you recover high-quality radiance fields from just a handful of images.
</p>
</td>
</tr> -->
<!-- Selected Awards and Honors -->
<table style="width:90%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody>
<br>
<heading>Selected Awards and Honors</heading>
<br>
<td style="padding:0px;width:100%;vertical-align:middle">
<p>
<li>2022-2024: Second Class Scholarship</li>
</p>
<p>
<li>2022-2024: Merit Student of Beijing Jiaotong University</li>
</p>
<p>
<li>Oct 2022: Second Prize of National College Student Mathematical Modeling Competition</li>
</p>
<p>
<li>2022-2023: Dean’s List </li>
</p>
<p>
<li>2022: Kaggle Sliver Medal (<strong>Top 5%</strong>): Feedback Prize - Evaluating Student Writing</li>
</p>
<!-- <p>
<li>2021: Zhongying Moral Education Scholarship (4000RMB¥)</li>
</p> -->
<!-- <p>
<li>2021: Merit Student of Peking University</li>
</p> -->
<!-- <p>
<li>2021: Ministry of Education Top Talent Program Scholarship (1000RMB¥)</li>
</p> -->
<!-- <p>
<li>2020: China University Computer Competition - Group Programming LadderCompetition Silver Medalist</li>
</p> -->
<!-- <p>
<li>2019: The 36th Chinese Physics Olympiad(CPHO) Silver Medalist</li>
</p>
<p>
<li>2019: The First Prize of National High School Mathematics Contest, Tianjin, China</li>
</p> -->
</td>
</tr>
</tbody></table>
<script type='text/javascript' id='mapmyvisitors' src='https://mapmyvisitors.com/map.js?cl=ffffff&w=a&t=m&d=N-HMxU5TSw_p9d8yn-JqcrusGHlPo92KV5_KtcGtXiI'></script>
<!-- Services -->
<!-- <table style="width:90%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody>
<tr>
<br>
<heading>Services</heading>
<br>
<td style="padding:0px;width:100%;vertical-align:middle">
<p>
<li>Conference Reviewer: ECCV2024</li>
</p>
<p>
<li>Journal Reviewer: Expert Systems with Applications</li>
</p>
</td>
</tr>
</tbody></table> -->
</tbody></table>
<table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody>
<tr>
<td style="padding:0px">
<br>
<p style="text-align:right;font-size:small;">
Feel free to steal this website's <a href="https://github.com/jonbarron/jonbarron_website">source code</a>.
</p>
</td>
</tr>
</tbody></table>
</td>
</tr>
</table>
</body>
</html>