forked from jonbarron/website
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
executable file
·190 lines (176 loc) · 11.2 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
<!DOCTYPE HTML>
<html lang="en"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
<title>Jiawei Zhang</title>
<meta name="author" content="Jiawei Zhang">
<meta name="viewport" content="width=device-width, initial-scale=1">
<link rel="stylesheet" type="text/css" href="stylesheet.css">
<link rel="icon" href="data:image/svg+xml,<svg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 120 90%22><text y=%22.9em%22 font-size=%2295%22>🍪</text></svg>">
</head>
<body>
<table style="width:100%;max-width:900px;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 style="text-align:center">
<name>Jiawei Zhang</name>
</p>
<p> I am currently pursuing my Ph.D. in Computer Science at the University of Chicago, where I serve as a research assistant in the Secure Learning Lab under the guidance of <a href="https://aisecure.github.io/">Prof. Bo Li</a>. Prior to this, I earned my Bachelor's degree from Zhejiang University.
</p>
<p>
My current research predominantly centers on <b>Trustworthy Machine Learning</b>, especially for Large Language Model (LLM). I’m particularly interested in enhancing their trustworthiness by mitigating issues like hallucination, using external knowledge sources as leverage. While my foundation in robustness, privacy, fairness, and explainability remains intact, my renewed focus aims at the integration of these principles into the development and understanding of LLMs, thereby ensuring they align more closely with human values and expectations.
</p>
<p style="text-align:center">
<a href="mailto:[email protected]">Email</a>  / 
<a href="data/Jiawei_CV.pdf">CV</a>  / 
<!-- <a href="data/Jiawei-bio.txt">Bio</a>  /  -->
<a href="https://scholar.google.com/citations?hl=en&user=vCY9ZRcAAAAJ">Google Scholar</a>  / 
<!-- <a href="https://twitter.com/jon_barron">Twitter</a>  /  -->
<a href="https://github.com/javyduck/">Github</a>
</p>
</td>
<td style="padding:2.5%;width:35%;max-width:40%">
<a href="images/jiawei.jpg"><img style="width:80%;max-width:80%" alt="profile photo" src="images/jiawei_circle.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">
<heading>News</heading>
<p>
<ul>
<li><b>[April, 2023]</b> I am co-hosting the <a href="https://trust-ai.github.io/SSAD2023/">Secure and Safe Autonomous Driving (SSAD) Workshop and Challenge</a> at CVPR 2023!</li>
<li><b>[May, 2023]</b> One paper accepted by <a href="https://www.usenix.org/conference/usenixsecurity23">USENIX Security 2023</a>.</li>
<li><b>[Feb, 2023]</b> I am selected as one of the CS candidates for Conference Presentation Awards for Graduate Students!</li>
<li><b>[Nov, 2022]</b> One paper accepted by <a href="https://satml.org/">IEEE SatML 2023</a>.</li>
<li><b>[Sept, 2022]</b> One paper accepted by <a href="https://nips.cc/Conferences/2022">NeurIPS 2022</a>.</li>
<li><b>[May, 2022]</b> One paper accepted by <a href="https://icml.cc/Conferences/2022">ICML 2022</a>.</li>
<li><b>[May, 2021]</b> One paper accepted by <a href="https://icml.cc/Conferences/2021">ICML 2021</a>.</li>
</ul>
</p>
</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">
<heading>Publications and Preprints</heading>
</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="samurai_stop()" onmouseover="samurai_start()">
<td style="padding:20px;width:31%;vertical-align:middle">
<div class="one">
<img src='images/diffsmooth.jpg' width="260" height="140">
</div>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://drive.google.com/file/d/1KPcm6oHooljYgGMke__O4EfJP7sqEUhU/view">
<papertitle>DiffSmooth: Certifiably Robust Learning via Diffusion Models and Local Smoothing</papertitle>
</a>
<br>
<b>Jiawei Zhang</b>, Zhongzhu Chen, Huan Zhang, Chaowei Xiao, Bo Li
<br>
<em>32th USENIX Security Symposium</em>, 2023
<p>
- We theoretically show that recovered instances by diffusion models are in the bounded neighborhood of the original instance with high probability; and the “one-shot" denoising diffusion probabilistic models (DDPM) can approximate the mean of the generated distribution of a continuous-time diffusion model. <br>
<font color="red">SOTA Certified Accuracy on ImageNet:</font> 77.2%, 63.2%, 53.0% under L2 radius 0.5, 1.0, 1.5, respectively. As a comparison, the current best results are 71.1%, 54.3%, and 38.1%.
</p>
</td>
</tr>
<tr onmouseout="samurai_stop()" onmouseover="samurai_start()">
<td style="padding:20px;width:31%;vertical-align:middle">
<div class="one">
<img src='images/mln_vi_before.jpg' width="260" height="160">
</div>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://arxiv.org/abs/2209.05055">
<papertitle>CARE: Certifiably Robust Learning with Reasoning via Variational Inference</papertitle>
</a>
<br>
<b>Jiawei Zhang</b>, Linyi Li, Ce Zhang, Bo Li
<br>
<em>IEEE Conference on Secure and Trustworthy Machine Learning</em>, 2023
<p>
- We propose a certifiably robust learning with reasoning pipeline (CARE), which consists of a learning component and a reasoning component. Concretely, we use a set of standard DNNs to serve as the learning component to make semantic predictions, and we leverage the probabilistic graphical models, such as Markov logic networks (MLN), to serve as the reasoning component to enable knowledge/logic reasoning.
</p>
</td>
</tr>
<tr onmouseout="samurai_stop()" onmouseover="samurai_start()">
<td style="padding:20px;width:31%;vertical-align:middle">
<div class="one">
<img src='images/sensing_reasoning.jpg' width="260" height="160">
</div>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://arxiv.org/abs/2003.00120">
<papertitle>Improving Certified Robustness via Statistical Learning with Logical Reasoning</papertitle>
</a>
<br>
Zhuolin Yang*, Zhikuan Zhao*, Boxin Wang, <b>Jiawei Zhang</b>, Linyi Li, Hengzhi Pei, Bojan Karlaš, Ji Liu, Heng Guo, Ce Zhang, Bo Li
<br>
<em>NeurIPS</em>, 2022
<p>
- This work provides the first knowledge-enabled, certifiably robust ML pipeline, sensing-reasoning pipeline, by combining statistical learning with logical reasoning. It first proves that the computational complexity of certifying the robustness of MLN is #P-hard and derives the first certified robustness bound for MLN by carefully analyzing different model regimes.
</p>
</td>
</tr>
<tr onmouseout="samurai_stop()" onmouseover="samurai_start()">
<td style="padding:20px;width:31%;vertical-align:middle">
<div class="one">
<img src='images/dsrs.jpg' width="260" height="160">
</div>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://arxiv.org/abs/2206.07912">
<papertitle>Double Sampling Randomized Smoothing</papertitle>
</a>
<br>
Linyi Li, <b>Jiawei Zhang</b>, Tao Xie, Bo Li
<br>
<em>ICML</em>, 2022
<p>
- This work exploits the sampled probability from an additional smoothing distribution to tighten the robustness certification of the previous smoothed classifier. Theoretically, under mild assumptions, we prove that DSRS can certify Θ(√d) robust radius under L2 norm where d is the input dimension, implying that DSRS may be able to break the curse of dimensionality of randomized smoothing.
</p>
</td>
</tr>
<tr onmouseout="samurai_stop()" onmouseover="samurai_start()">
<td style="padding:20px;width:31%;vertical-align:middle">
<div class="one">
<img src='images/psba.jpg' width="260" height="160">
</div>
</td>
<td style="padding:20px;width:75%;vertical-align:middle">
<a href="https://arxiv.org/abs/2106.06056">
<papertitle>Progressive-scale boundary blackbox attack via projective gradient estimation</papertitle>
</a>
<br>
<b>Jiawei Zhang*</b>, Linyi Li*, Huichen Li, Xiaolu Zhang, Shuang Yang, Bo Li
<br>
<em>ICML</em>, 2021
<p>
- The first theoretical framework to analyze boundary blackbox attacks with general projection functions and characterize the key characteristics and trade-offs for a good projective gradient estimator. Based on this, we propose Progressive-Scale based projective Boundary Attack (PSBA) via progressively searching for the optimal scale in a self-adaptive way under spatial, frequency, and spectrum scales.
</p>
</td>
</tr>
</tbody></table>
<table width="100%" align="center" border="0" cellspacing="0" cellpadding="20">
<tbody><tr>
<td>
<br>
<p align="right">
<font size="2">
Template from <a href="https://jonbarron.info/">Jon Barron</a>.
</font>
</p>
</td>
</tr>
</tbody></table>
</a></td>
</tr>
</tbody></table>
</html>