-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
306 lines (287 loc) · 18.3 KB
/
Copy pathindex.html
File metadata and controls
306 lines (287 loc) · 18.3 KB
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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="description" content="Computational Sensing and Information Processing Lab at UC Riverside. Research in computational imaging, inverse problems, and robust and trustworthy machine learning.">
<meta name="author" content="M. Salman Asif">
<title>CSIPLab | Computational Sensing and Information Processing Lab, UCR</title>
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Lato:ital,wght@0,400;0,700;1,400&display=swap" rel="stylesheet">
<link rel="stylesheet" href="src/css/home.css">
</head>
<body>
<div class="page">
<header class="site-header">
<h1 class="lab-name">Computational Sensing and Information Processing Lab</h1>
<p class="lab-tagline">Computational Imaging and Machine Learning at UC Riverside</p>
<p class="lab-intro">
We are a research group in the
<a href="https://www.ece.ucr.edu/" target="_blank" rel="noopener">Department of Electrical and Computer Engineering</a>
at the University of California, Riverside, directed by
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">M. Salman Asif</a>.
Our work spans computational imaging and optics, inverse problems, and machine learning —
with a recent focus on wavefront sensing, robust multimodal learning, and trustworthy AI.
</p>
<p class="inline-links">
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">Director</a>
<span class="sep">/</span>
<a href="team.html">Team</a>
<span class="sep">/</span>
<a href="https://github.com/CSIPlab" target="_blank" rel="noopener">GitHub</a>
<span class="sep">/</span>
<a href="https://www.ece.ucr.edu/" target="_blank" rel="noopener">UCR ECE</a>
</p>
</header>
<main>
<h2 class="section-title">Research</h2>
<p class="section-note">Recent work is <span class="paper-tag">highlighted</span>.</p>
<!--
To add a project, copy one <article class="paper"> block below and place it in
publication-date order (newest at the top). Add the class "paper--new" for
highlighting, and remove it from older entries as new ones are added.
Thumbnails are served from each project's own sub-directory.
-->
<div class="papers">
<!-- ECCV 2026 · project page lives in the separate CSIPlab/coprime-psi repository -->
<article class="paper paper--new">
<div class="paper-thumb">
<a href="https://csiplab.github.io/coprime-psi/" target="_blank" rel="noopener">
<img src="https://csiplab.github.io/coprime-psi/assets/intro_figure_v5.png"
alt="Self-reference interferometry wavefront sensing overview">
</a>
</div>
<div class="paper-body">
<a class="paper-title" href="https://csiplab.github.io/coprime-psi/" target="_blank" rel="noopener">Provable and Robust Wavefront Sensing via Self-Reference Interferometry</a>
<span class="paper-tag">(New)</span>
<br>
<span class="paper-authors">
<a href="https://nebiyouten.github.io/" target="_blank" rel="noopener">Nebiyou Yismaw</a>,
<a href="https://vishwa91.github.io/" target="_blank" rel="noopener">Vishwanath Saragadam</a>,
<a href="https://www.ece.cmu.edu/directory/bios/sankaranarayanan-aswin.html" target="_blank" rel="noopener">Aswin C. Sankaranarayanan</a>,
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">M. Salman Asif</a>
</span>
<br>
<span class="paper-venue"><em>European Conference on Computer Vision (ECCV)</em>, 2026</span>
<span class="paper-award">(Oral Presentation)</span>
<br>
<span class="paper-links">
<a href="https://csiplab.github.io/coprime-psi/" target="_blank" rel="noopener">project page</a>
<span class="sep">/</span>
<a href="https://arxiv.org/abs/2604.03564" target="_blank" rel="noopener">arXiv</a>
<span class="sep">/</span>
<a href="https://csiplab.github.io/coprime-psi/assets/paper.pdf" target="_blank" rel="noopener">paper</a>
<span class="sep">/</span>
<a href="https://csiplab.github.io/coprime-psi/assets/supplementary.pdf" target="_blank" rel="noopener">supplementary</a>
</span>
<p class="paper-tldr">
Interfering an incoming wave with spatially shifted copies of itself removes the need for a stable
reference beam. We recover the global phase by propagating pairwise differences across a connected
graph, and prove that co-prime shifts guarantee connectivity and bound error accumulation —
recovering complete phase profiles from as few as eight measurements, validated on a hardware prototype.
</p>
</div>
</article>
<article class="paper">
<div class="paper-thumb">
<a href="cross-modal-proxy-tokens/">
<img loading="lazy" src="./cross-modal-proxy-tokens/img/model-diagram-4.png" alt="Cross-Modal Proxy Tokens">
</a>
</div>
<div class="paper-body">
<a class="paper-title" href="cross-modal-proxy-tokens/">Robust Multimodal Learning via Cross-Modal Proxy Tokens</a>
<br>
<span class="paper-authors">
<a href="https://kaykobad.github.io/" target="_blank" rel="noopener">Md Kaykobad Reza</a>,
Ameya Patil,
Mashhour Solh,
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">M. Salman Asif</a>
</span>
<br>
<span class="paper-venue"><em>Transactions on Machine Learning Research (TMLR)</em>, 2025</span>
<br>
<span class="paper-links">
<a href="cross-modal-proxy-tokens/">project page</a>
<span class="sep">/</span>
<a href="https://openreview.net/forum?id=Wtc6wvcYJ0" target="_blank" rel="noopener">OpenReview</a>
<span class="sep">/</span>
<a href="https://arxiv.org/abs/2501.17823" target="_blank" rel="noopener">arXiv</a>
<span class="sep">/</span>
<a href="https://github.com/CSIPlab/Cross-Modal-Proxy-Tokens" target="_blank" rel="noopener">code</a>
</span>
<p class="paper-tldr">
Cross-modal proxy tokens approximate the class token of a missing modality by attending only to the
tokens of the available modality — no modality generation, no auxiliary networks. With rank-1
LoRA adapters in frozen unimodal encoders, this beats state-of-the-art baselines across missing rates
on five datasets while staying competitive with complete modalities.
</p>
</div>
</article>
<article class="paper">
<div class="paper-thumb">
<a href="icip2025_multimodal/">
<img loading="lazy" src="./icip2025_multimodal/img/mml-tutorial.png" alt="Robust Multimodal Learning Tutorial">
</a>
</div>
<div class="paper-body">
<a class="paper-title" href="icip2025_multimodal/">Foundations and Recent Trends in Robust Multimodal Learning</a>
<br>
<span class="paper-authors">
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">M. Salman Asif</a>,
<a href="https://kaykobad.github.io/" target="_blank" rel="noopener">Md Kaykobad Reza</a>
</span>
<br>
<span class="paper-venue"><em>IEEE International Conference on Image Processing (ICIP)</em>, 2025</span>
<span class="paper-award">(Tutorial)</span>
<br>
<span class="paper-links">
<a href="icip2025_multimodal/">tutorial page</a>
</span>
<p class="paper-tldr">
A tutorial overview of robust multimodal learning, from the fundamentals of fusion, alignment, and
representation learning through the architectures and models that keep performance from collapsing
when modalities are noisy or missing.
</p>
</div>
</article>
<article class="paper">
<div class="paper-thumb">
<a href="slug/">
<img loading="lazy" src="./slug/assets/project/slug/slug-framework.svg" alt="SLUG: Single Layer Unlearning Gradient">
</a>
</div>
<div class="paper-body">
<a class="paper-title" href="slug/">Targeted Unlearning with Single Layer Unlearning Gradient</a>
<br>
<span class="paper-authors">
<a href="https://zikuicai.github.io/" target="_blank" rel="noopener">Zikui Cai</a>,
<a href="https://ytengtan.github.io/" target="_blank" rel="noopener">Yaoteng Tan</a>,
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">M. Salman Asif</a>
</span>
<br>
<span class="paper-venue"><em>International Conference on Machine Learning (ICML)</em>, 2025</span>
<br>
<span class="paper-links">
<a href="slug/">project page</a>
<span class="sep">/</span>
<a href="https://arxiv.org/abs/2407.11867" target="_blank" rel="noopener">arXiv</a>
<span class="sep">/</span>
<a href="https://github.com/CSIPlab/SLUG" target="_blank" rel="noopener">code</a>
<span class="sep">/</span>
<a href="https://colab.research.google.com/drive/1sjLWKPIXi961KPV-t1ugIdwaF7bdBjt7?usp=sharing" target="_blank" rel="noopener">demo</a>
</span>
<p class="paper-tldr">
SLUG unlearns targeted information by updating a single critical layer from a one-time gradient
computation, using layer-importance and gradient-alignment metrics to pick that layer. It removes
concrete and abstract concepts from CLIP, Stable Diffusion, and VLMs at a small fraction of the
compute of existing methods.
</p>
</div>
</article>
<article class="paper">
<div class="paper-thumb">
<a href="Missing-Modality-Adaptation/">
<img loading="lazy" src="./Missing-Modality-Adaptation/img/ssf-final.png" alt="Parameter-efficient adaptation for missing modalities">
</a>
</div>
<div class="paper-body">
<a class="paper-title" href="Missing-Modality-Adaptation/">Robust Multimodal Learning with Missing Modalities via Parameter-Efficient Adaptation</a>
<br>
<span class="paper-authors">
<a href="https://kaykobad.github.io/" target="_blank" rel="noopener">Md Kaykobad Reza</a>,
Ashley Prater-Bennette,
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">M. Salman Asif</a>
</span>
<br>
<span class="paper-venue"><em>IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)</em>, 2024</span>
<br>
<span class="paper-links">
<a href="Missing-Modality-Adaptation/">project page</a>
<span class="sep">/</span>
<a href="https://ieeexplore.ieee.org/document/10713849" target="_blank" rel="noopener">IEEE Xplore</a>
<span class="sep">/</span>
<a href="https://arxiv.org/abs/2310.03986" target="_blank" rel="noopener">arXiv</a>
<span class="sep">/</span>
<a href="https://github.com/CSIPlab/Robust-multimodal-learning" target="_blank" rel="noopener">code</a>
</span>
<p class="paper-tldr">
Pretrained multimodal networks degrade sharply when a modality is absent at test time. Modulating
intermediate features with fewer than 1% of the model's parameters compensates for the missing inputs,
and often beats dedicated networks trained for each available modality combination — demonstrated
on five tasks across seven datasets.
</p>
</div>
</article>
<article class="paper">
<div class="paper-thumb">
<a href="MMSFormer/">
<img loading="lazy" src="./MMSFormer/img/MMSFormer-min.png" alt="MMSFormer architecture">
</a>
</div>
<div class="paper-body">
<a class="paper-title" href="MMSFormer/">MMSFormer: Multimodal Transformer for Material and Semantic Segmentation</a>
<br>
<span class="paper-authors">
<a href="https://kaykobad.github.io/" target="_blank" rel="noopener">Md Kaykobad Reza</a>,
Ashley Prater-Bennette,
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">M. Salman Asif</a>
</span>
<br>
<span class="paper-venue"><em>IEEE Open Journal of Signal Processing (OJSP)</em>, 2024</span>
<br>
<span class="paper-links">
<a href="MMSFormer/">project page</a>
<span class="sep">/</span>
<a href="https://ieeexplore.ieee.org/document/10502124" target="_blank" rel="noopener">IEEE Xplore</a>
<span class="sep">/</span>
<a href="https://arxiv.org/abs/2309.04001" target="_blank" rel="noopener">arXiv</a>
<span class="sep">/</span>
<a href="https://github.com/CSIPlab/MMSFormer" target="_blank" rel="noopener">code</a>
</span>
<p class="paper-tldr">
A mixed-modality fusion block that combines information from arbitrary modality combinations, built
into a Multi-Modal Segmentation TransFormer for material and semantic segmentation. Accuracy improves
monotonically as modalities are added, setting state of the art on three datasets.
</p>
</div>
</article>
<article class="paper">
<div class="paper-thumb">
<a href="DomainExpansionInverseProblems/">
<img loading="lazy" src="./DomainExpansionInverseProblems/static/images/intro_figure.png" alt="Domain expansion for inverse problems">
</a>
</div>
<div class="paper-body">
<a class="paper-title" href="DomainExpansionInverseProblems/">Domain Expansion via Network Adaptation for Solving Inverse Problems</a>
<br>
<span class="paper-authors">
<a href="https://nebiyouten.github.io/" target="_blank" rel="noopener">Nebiyou Yismaw</a>,
<a href="https://ukmlv.github.io/" target="_blank" rel="noopener">Ulugbek S. Kamilov</a>,
<a href="https://intra.ece.ucr.edu/~sasif/" target="_blank" rel="noopener">M. Salman Asif</a>
</span>
<br>
<span class="paper-venue"><em>IEEE Transactions on Computational Imaging (TCI)</em>, 2024</span>
<br>
<span class="paper-links">
<a href="DomainExpansionInverseProblems/">project page</a>
<span class="sep">/</span>
<a href="https://doi.org/10.1109/TCI.2024.3377101" target="_blank" rel="noopener">IEEE Xplore</a>
<span class="sep">/</span>
<a href="https://arxiv.org/abs/2310.06235" target="_blank" rel="noopener">arXiv</a>
</span>
<p class="paper-tldr">
Networks trained to solve inverse problems are fragile under shifts in data distribution, measurement
operator, and noise level. We characterize the effects of those shifts and propose a flexible,
parameter-efficient framework that adapts a pretrained network to expand its domain of validity.
</p>
</div>
</article>
</div>
</main>
<footer class="site-footer">
<p>© 2023–2026 Computational Sensing and Information Processing Lab, University of California, Riverside</p>
</footer>
</div>
</body>
</html>