-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathresearch.html
More file actions
644 lines (559 loc) · 22.9 KB
/
research.html
File metadata and controls
644 lines (559 loc) · 22.9 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
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
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="icon" href="logo/log_logo.png" type="image/png">
<title>Research | Lee Optimization Group</title>
<link href="https://fonts.googleapis.com/css2?family=EB+Garamond:ital,wght@0,400;0,500;1,400&family=Inter:wght@300;400;500;600&display=swap" rel="stylesheet">
<style>
*, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
html {
overflow-y: scroll;
}
body {
font-family: 'Inter', system-ui, sans-serif;
color: #111;
background: #ffffff;
min-height: 100vh;
}
.inner {
max-width: 800px;
margin: 0 auto;
padding: clamp(24px, 5vw, 48px) clamp(16px, 4vw, 32px) clamp(36px, 5vw, 60px);
}
header {
padding-bottom: 28px;
margin-bottom: 40px;
border-bottom: 1px solid #e2e6f0;
}
.lab-name {
font-family: 'EB Garamond', Georgia, serif;
font-size: 2.8em;
font-weight: 500;
color: #111;
letter-spacing: 0;
line-height: 0.96;
display: block;
margin-bottom: 18px;
text-decoration: none;
}
.lab-name:hover {
text-decoration: none;
color: #111;
}
nav {
display: flex;
flex-wrap: wrap;
align-items: center;
gap: clamp(10px, 3vw, 24px);
}
nav a {
text-decoration: none;
color: #6b7280;
font-size: 1em;
font-weight: 400;
letter-spacing: 0.01em;
padding-bottom: 2px;
border-bottom: 1.5px dashed transparent;
transition: color 0.15s, border-color 0.15s;
}
nav a:hover {
color: #1B3A8A;
border-bottom: 1.5px solid #1B3A8A;
text-decoration: none;
}
nav a.active {
color: #111;
font-weight: 600;
border-bottom: 1.5px solid #111;
}
.nav-drawer a.active { color: #111; font-weight: 600; }
a { color: #1B3A8A; text-decoration: none; }
a:hover { text-decoration: underline; }
h2 {
font-family: 'Inter', system-ui, sans-serif;
font-size: 1.20em;
font-weight: 700;
letter-spacing: 0.01em;
text-transform: none;
color: #111;
margin-bottom: 28px;
}
.research-intro {
margin-bottom: 26px;
}
p {
font-size: 0.93em;
line-height: 1.75;
color: #374151;
margin-bottom: 0;
}
.fold-trigger {
align-items: baseline;
background: none;
border: none;
color: #6b7280;
cursor: pointer;
display: inline-flex;
font-size: 0.82em;
font-weight: 500;
gap: 6px;
margin-left: 4px;
padding: 0;
}
.fold-trigger:hover {
color: #1B3A8A;
}
.fold-trigger:hover .fold-label {
text-decoration: underline;
text-underline-offset: 3px;
}
.fold-trigger:focus {
outline: none;
}
.fold-trigger:focus-visible {
outline: 1px solid currentColor;
outline-offset: 3px;
}
.fold-caret {
font-size: 0.85em;
transition: transform 0.15s ease;
}
.fold-trigger[aria-expanded="true"] .fold-caret {
transform: rotate(90deg);
}
.fold-detail {
border-left: 2px solid #e2e8f4;
color: #4b5563;
display: none;
font-size: 0.9em;
line-height: 1.68;
margin-top: 10px;
padding-left: 12px;
}
.fold-detail.open {
display: block;
}
.fold-item {
margin-bottom: 9px;
}
.fold-item:last-child {
margin-bottom: 0;
}
.fold-item-title {
color: #111;
display: block;
font-weight: 600;
margin-bottom: 2px;
}
.paper-tag {
color: #8a9099;
font-size: 0.82em;
font-weight: 400;
}
.paper-tag a {
color: inherit;
font-weight: 500;
position: relative;
text-decoration: none;
transition: color 0.15s ease;
}
.paper-tag a::after {
border-bottom: 1px dotted currentColor;
bottom: -2px;
content: "";
left: 0;
position: absolute;
transform: scaleX(0);
transform-origin: left;
transition: transform 0.18s ease;
width: 100%;
}
.paper-tag a:hover,
.paper-tag a:focus {
color: #1B3A8A;
outline: none;
text-decoration: none;
}
.paper-tag a:hover::after,
.paper-tag a:focus::after {
transform: scaleX(1);
transition-duration: 0s;
}
.section {
display: grid;
grid-template-columns: 165px minmax(0, 1fr);
align-items: start;
column-gap: 22px;
row-gap: 10px;
margin-bottom: 40px;
}
.section .section-text {
min-width: 0;
}
.section img {
width: 165px;
max-width: none;
height: auto;
object-fit: contain;
border: 1.5px solid #b8bec8;
border-radius: 6px;
margin-top: 4px;
}
.section.wrap-flow {
display: block;
}
.section.wrap-flow::after {
content: "";
display: block;
clear: both;
}
.section.wrap-flow img {
float: left;
margin: 4px 22px 10px 0;
}
.section.wrap-flow .section-text {
display: block;
}
.section-title { font-weight: 700; }
.red { color: #d64541; font-weight: 600; }
.orange { color: #f39c12; font-weight: 600; }
.green { color: #27ae60; font-weight: 600; }
.blue { color: #2e6bd9; font-weight: 600; }
footer {
margin-top: 52px;
padding-top: 20px;
border-top: 1px solid #e2e6f0;
display: flex;
align-items: center;
justify-content: space-between;
font-size: 0.75em;
color: #6b7280;
}
.footer-logos {
display: flex;
align-items: center;
gap: 14px;
}
.footer-logos img {
height: 28px;
width: auto;
object-fit: contain;
opacity: 1;
}
.collab-logos {
display: flex;
flex-wrap: wrap;
align-items: center;
justify-content: center;
column-gap: 16px;
row-gap: 12px;
margin-top: 20px;
margin-left: auto;
margin-right: auto;
background: #fff;
border-radius: 6px;
border: 1px solid #e8e4dd;
padding: 16px 18px;
width: fit-content;
}
.collab-logos img {
height: 40px;
width: auto;
object-fit: contain;
}
.menu-toggle {
display: none;
background: #ffffff;
border: 1px solid #e2e6f0;
border-radius: 6px;
font-size: 1.3em;
cursor: pointer;
color: #374151;
padding: 4px 8px;
line-height: 1;
}
@media (max-width: 900px) {
.inner {
padding: 28px 18px 40px;
}
header {
padding-bottom: 4px;
margin-bottom: 16px;
display: grid;
grid-template-columns: minmax(0, 1fr) auto;
align-items: start;
column-gap: 10px;
}
.lab-name {
min-width: 0;
}
.menu-toggle {
display: block;
}
nav {
display: none;
}
.footer-logos img {
height: 18px;
}
}
/* Drawer */
.nav-overlay {
display: none;
position: fixed;
inset: 0;
background: rgba(0,0,0,0.3);
z-index: 999;
}
.nav-drawer {
position: fixed;
top: 0;
right: -280px;
width: 240px;
height: 100vh;
background: #ffffff;
box-shadow: -4px 0 24px rgba(0,0,0,0.12);
z-index: 1000;
transition: right 0.3s ease;
padding: 64px 32px 40px;
display: flex;
flex-direction: column;
gap: 20px;
}
.nav-drawer.open {
right: 0;
}
.nav-overlay.open {
display: block;
}
.drawer-close {
position: absolute;
top: 20px;
right: 20px;
background: none;
border: none;
font-size: 1.4em;
cursor: pointer;
color: #374151;
line-height: 1;
}
.nav-drawer a {
text-decoration: none;
color: #374151;
font-size: 1.05em;
font-weight: 400;
}
.nav-drawer a:hover {
color: #1B3A8A;
}
</style>
</head>
<body>
<div class="inner">
<header>
<a href="index.html" class="lab-name">Lee Optimization Group</a>
<button class="menu-toggle" aria-label="Menu">☰</button>
<nav>
<a href="index.html">Home</a>
<a href="research.html" class="active">Research</a>
<a href="publications.html">Publications</a>
<a href="group.html">Members</a>
<a href="hiring.html">Hiring</a>
<a href="contact.html">Contact</a>
</nav>
</header>
<main>
<h2>Research Focus</h2>
<div class="section">
<img src="research_focus/opt-compression.svg" alt="Model Compression">
<div class="section-text">
<p>
<span class="red">Model Compression</span>
As AI models become larger and more computationally demanding, compression is essential for making them practical and sustainable.
We develop optimization methods that reduce training and inference costs while preserving model performance.
Our work focuses on extreme compression across the AI model lifecycle.
<button class="fold-trigger" type="button" aria-expanded="false" aria-controls="compressionDetail">
<span class="fold-label">more</span><span class="fold-caret" aria-hidden="true">🔍</span>
</button>
</p>
<div class="fold-detail" id="compressionDetail">
<div class="fold-item">
<span class="fold-item-title">Extreme Sparsity <span class="paper-tag">(<a href="https://arxiv.org/abs/2510.01650" target="_blank">ICLR '26</a>, <a href="https://arxiv.org/abs/2506.06866" target="_blank">ICML '25</a>, <a href="https://arxiv.org/abs/2406.15524" target="_blank">EMNLP '24</a>)</span></span>
We push the limits of LLM sparsity through advanced optimization frameworks such as ADMM, aiming for robust generalization even at extreme compression levels.
</div>
<div class="fold-item">
<span class="fold-item-title">Joint Compression</span>
We develop unified frameworks that integrate pruning, quantization, and distillation to reduce error accumulation and preserve core model capabilities, including reasoning, in highly compressed states.
</div>
</div>
</div>
</div>
<div class="section">
<img src="research_focus/opt-dist.png" alt="Distributed Training">
<div class="section-text">
<p>
<span class="green">Distributed Training</span>
The shift toward massive-scale models has made distributed training essential for workloads that no single machine can support.
From an optimization perspective, the key challenge is maintaining efficiency while managing the overhead of communication between devices.
Our research develops robust algorithms for decentralized and high-latency training environments.
<button class="fold-trigger" type="button" aria-expanded="false" aria-controls="distributedDetail">
<span class="fold-label">more</span><span class="fold-caret" aria-hidden="true">🔍</span>
</button>
</p>
<div class="fold-detail" id="distributedDetail">
<div class="fold-item">
<span class="fold-item-title">Communication-Efficient Training <span class="paper-tag">(<a href="https://arxiv.org/abs/2602.18181" target="_blank">arXiv '26</a>)</span></span>
We design methods that reduce the heavy data-sharing requirements of distributed training through information compression, reduced synchronization, and zeroth-order optimization.
</div>
<div class="fold-item">
<span class="fold-item-title">Asynchronous Training <span class="paper-tag">(<a href="https://arxiv.org/abs/2602.03515" target="_blank">arXiv '26</a>)</span></span>
We create algorithms that allow devices to work independently and reduce waiting time, with a focus on correcting errors caused by slightly outdated information.
</div>
<div class="fold-item">
<span class="fold-item-title">Device-Heterogeneous Training</span>
We develop adaptive strategies for imbalanced computational resources across devices, aiming for consistent global convergence despite variations in hardware performance.
</div>
</div>
</div>
</div>
<div class="section">
<img src="research_focus/opt-applied.png" alt="Deep Learning Applications">
<div class="section-text">
<p>
<span class="blue">Deep Learning Applications</span>
Optimization is not only a subject of study in itself, but also a versatile lens through which we tackle challenges in interpretability, uncertainty quantification, continual learning, and other real-world deep learning systems.
Our research explores how optimization principles can be extended and applied to address such challenges.
<button class="fold-trigger" type="button" aria-expanded="false" aria-controls="applicationsDetail">
<span class="fold-label">more</span><span class="fold-caret" aria-hidden="true">🔍</span>
</button>
</p>
<div class="fold-detail" id="applicationsDetail">
<div class="fold-item">
<span class="fold-item-title">Interpretability <span class="paper-tag">(<a href="https://arxiv.org/abs/2505.16705" target="_blank">NeurIPS '25</a>, <a href="https://arxiv.org/abs/2302.14260" target="_blank">ICML '23</a>)</span></span>
We bridge complex model architectures and human understanding, from uncovering model logic in realistic scenarios to evaluating the reliability of model outputs.
</div>
<div class="fold-item">
<span class="fold-item-title">Continual Adaptation <span class="paper-tag">(<a href="https://arxiv.org/abs/2510.25798" target="_blank">NeurIPS '25</a>)</span></span>
We study constrained optimization problems where models adapt to new knowledge without disrupting what has already been learned, opening principled approaches to continual learning.
</div>
</div>
</div>
</div>
<div class="section">
<img src="research_focus/opt-deep.jpeg" alt="Advanced Optimization">
<div class="section-text">
<p>
<span class="orange">Advanced Optimization</span>
Optimization has long been a source of crucial ideas that drastically enhance all corners of deep neural network training, and many of its most impactful questions remain open.
Our research investigates these questions and develops optimization principles and algorithms for modern deep learning systems.
<button class="fold-trigger" type="button" aria-expanded="false" aria-controls="optimizationDetail">
<span class="fold-label">more</span><span class="fold-caret" aria-hidden="true">🔍</span>
</button>
</p>
<div class="fold-detail" id="optimizationDetail">
<div class="fold-item">
<span class="fold-item-title">Optimization for Generalization <span class="paper-tag">(<a href="https://arxiv.org/abs/2311.17539" target="_blank">UAI '25</a>, <a href="https://arxiv.org/abs/2502.18153" target="_blank">ICML '25</a>)</span></span>
We study the role of loss landscape curvature in generalization and develop flatness-oriented optimization strategies with both theoretical advantages and scalability for modern deep neural networks.
</div>
<div class="fold-item">
<span class="fold-item-title">Zeroth-Order Optimization for Black-Box Model Training <span class="paper-tag">(<a href="https://arxiv.org/abs/2504.06838" target="_blank">ICLR '25</a>)</span></span>
We develop methods that operate without direct gradient access, motivated by settings such as proprietary model APIs and decentralized learning systems where gradients are unavailable.
</div>
</div>
</div>
</div>
<h2>Collaborators</h2>
<p>We actively collaborate with leading research groups worldwide.</p>
<div class="collab-logos">
<img src="logo/ista_logo.png" alt="ISTA" style="height:36px;">
<img src="logo/epfl_logo.png" alt="EPFL" style="height:16px;">
<img src="logo/oxford_logo.png" alt="Oxford" style="height:24px;">
<img src="logo/anu_logo.png" alt="ANU" style="height:24px;">
<img src="logo/cmu.logo.png" alt="CMU" style="height:26px;">
<img src="logo/google_logo.svg" alt="Google" style="height:28px;">
<img src="logo/etri_logo.png" alt="ETRI" style="height:28px;">
</div>
</main>
<footer>
<span>© 2026 LOG. All rights reserved.</span>
<div class="footer-logos">
<a href="https://www.postech.ac.kr" target="_blank"><img src="logo/postech_logo.png" alt="POSTECH"></a>
<a href="https://ai.postech.ac.kr" target="_blank"><img src="logo/gsai_logo.png" alt="GSAI"></a>
<a href="https://cse.postech.ac.kr" target="_blank"><img src="logo/cs_logo.png" alt="CS"></a>
<img src="logo/log_logo.png" alt="LOG">
</div>
</footer>
</div>
<div class="nav-overlay" id="navOverlay"></div>
<div class="nav-drawer" id="navDrawer">
<button class="drawer-close" id="drawerClose">✕</button>
<a href="index.html">Home</a>
<a href="research.html" class="active">Research</a>
<a href="publications.html">Publications</a>
<a href="group.html">Members</a>
<a href="hiring.html">Hiring</a>
<a href="contact.html">Contact</a>
</div>
<script>
var toggle = document.querySelector('.menu-toggle');
var drawer = document.getElementById('navDrawer');
var overlay = document.getElementById('navOverlay');
var close = document.getElementById('drawerClose');
function openDrawer() {
drawer.classList.add('open');
overlay.classList.add('open');
}
function closeDrawer() {
drawer.classList.remove('open');
overlay.classList.remove('open');
}
toggle.addEventListener('click', function() { drawer.classList.contains('open') ? closeDrawer() : openDrawer(); });
close.addEventListener('click', closeDrawer);
overlay.addEventListener('click', closeDrawer);
document.querySelectorAll('.fold-trigger').forEach(function(button) {
button.addEventListener('click', function() {
var detail = document.getElementById(button.getAttribute('aria-controls'));
var isOpen = button.getAttribute('aria-expanded') === 'true';
button.setAttribute('aria-expanded', String(!isOpen));
if (detail) {
detail.classList.toggle('open', !isOpen);
}
});
});
function applyConditionalResearchWrap() {
var sections = document.querySelectorAll('.section');
sections.forEach(function(section) {
var image = section.querySelector('img');
var textBlock = section.querySelector('.section-text p');
if (!image || !textBlock) {
return;
}
section.classList.remove('wrap-flow');
var textStyle = window.getComputedStyle(textBlock);
var lineHeight = parseFloat(textStyle.lineHeight);
if (!lineHeight || Number.isNaN(lineHeight)) {
lineHeight = parseFloat(textStyle.fontSize) * 1.75;
}
var textHeight = textBlock.getBoundingClientRect().height;
var imageHeight = image.getBoundingClientRect().height;
var threshold = imageHeight + (2 * lineHeight);
if (textHeight > threshold) {
section.classList.add('wrap-flow');
}
});
}
var wrapResizeTimer;
function onWrapResize() {
window.clearTimeout(wrapResizeTimer);
wrapResizeTimer = window.setTimeout(applyConditionalResearchWrap, 80);
}
window.addEventListener('load', applyConditionalResearchWrap);
window.addEventListener('resize', onWrapResize);
if (document.fonts && document.fonts.ready) {
document.fonts.ready.then(applyConditionalResearchWrap);
}
</script>
</body>
</html>