-
Notifications
You must be signed in to change notification settings - Fork 0
/
niivueHackathone2023.html
484 lines (416 loc) · 18 KB
/
niivueHackathone2023.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
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
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no">
<link href="css/fontawesome-free-6.2.1-web/css/all.css" rel="stylesheet">
<script src="lib/colorbrewer.v1.min.js" charset="utf-8"></script>
<script src="lib/colorStringStandalone.js" charset="utf-8"></script>
<script type="text/javascript" src="lib/jquery-2.2.4.min.js"></script>
<title>Neuroimaging On The Edge (NOTE)</title>
<meta name="description" content="University of South Carolina Hackathon">
<meta name="author" content="Sergey M Plis">
<meta name="apple-mobile-web-app-capable" content="yes">
<meta name="apple-mobile-web-app-status-bar-style" content="black-translucent">
<link rel="stylesheet" href="dist/reset.css">
<link rel="stylesheet" href="dist/reveal.css">
<!-- Code syntax highlighting -->
<link rel="stylesheet" href="plugin/highlight/monokai.css" id="highlight-theme">
<!-- <link rel="stylesheet" href="lib/css/zenburn.css"> -->
<link rel="stylesheet" href="css/custom.css">
<link rel="stylesheet" href="dist/theme/aml.css" id="theme">
<!-- Printing and PDF exports -->
<script>
var link = document.createElement( 'link' );
link.rel = 'stylesheet';
link.type = 'text/css';
link.href = window.location.search.match( /print-pdf/gi ) ? 'css/print/pdf.css' : 'css/print/paper.scss';
document.getElementsByTagName( 'head' )[0].appendChild( link );
</script>
</head>
<body>
<div class="reveal">
<!-- In between the <div="reveal"> and the <div class="slides">-->
<!-- <header style="position: absolute; top: 10px; left: 100px; z-index: 500; font-size:100px;background-color: rgba(0,0,0,0); text-align: center !important"></header> -->
<!-- In between the <div="reveal"> and the <div class="slides">-->
<!-- Any section element inside of this container is displayed as a slide -->
<div class="slides">
<section>
<section>
<p>
<h2>AI <alert>NOTE</alert>: AI <alert>N</alert>euroimaging <alert>O</alert>n <alert>T</alert>he <alert>E</alert>dge</h2>
<h4><i class="fa-solid fa-at"></i> Edge Machine Learning for Brain Imaging Hackathon 2023</h4>
<p>
Sergey Plis, PhD
<p><small>
Associate Professor, CS GSU
</small>
<p>
<p>
<p>
<p>
</section>
</section>
<section>
<section>
<h1>Vision</h1>
</section>
<section>
<h2>Scaling Laws</h2>
<row style="width:150%; margin-left: -220px">
<col50>
<img src="figures/FLOPs_scale.png" alt="FLOPS scale" style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1); margin-bottom: 120px; margin-top: -30px; margin-left: 0px; width: 100%" class="stretch">
</col50>
<col50>
<img src="figures/Params_scale.png" alt="Params scale" style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1); margin-bottom: 10%; margin-top: -10px; width: 100%" class="stretch">
</col50>
</row>
<div class='slide-footer' style="text-align: left;">
<a href="https://arxiv.org/abs/2005.14165">Language Models are Few-Shot Learners</a> 2020<br>
<a href="https://arxiv.org/abs/1712.00409">Deep Learning Scaling is Predictable, Empirically</a> 2017
</div>
</section>
<section data-background="figures/foundational_parameters.png" data-background-size="contain" data-vertical-align-top>
<h2 style="text-shadow: 10px 10px 10px #002b36; color: #fff3e3;">Foundational Models</h2>
</section>
<section data-background="figures/Corporate_foundational.png" data-background-size="contain" data-vertical-align-top>
<h2 style="text-shadow: 10px 10px 10px #002b36; color: #fff3e3;">Corporate Controlled AI</h2>
</section>
<section>
<h2>But what about privacy?</h2>
</section>
<section data-background="figures/COINSTAC_vision.png" data-background-size="contain" data-vertical-align-top>
<h2>coinstac.org</h2>
</section>
<section>
<h2>coinstac.org</h2>
Bring Computation to the user
</section>
<section>
<h2>AI NOTE</h2>
Bring AI models to the user <em>in their Browser</em>
</section>
<section>
<h2>Problems to be addressed</h2>
<ul>
<li class="fragment roll-in"> Segmentation
<li class="fragment roll-in"> Super-Resolution
<li class="fragment roll-in"> Synthetic MRI generation
<li class="fragment roll-in"> Tumor segmentation
<li class="fragment roll-in"> Surface Generation
</ul>
</section>
<section>
<h2>Essential model traits</h2>
<ul>
<li class="fragment roll-in"> Small number of weights
<li class="fragment roll-in"> Minimal memory footprint
<li class="fragment roll-in"> Powerful induction bias
<li class="fragment roll-in"> Transparency and interpretability
</ul>
</section>
</section>
<section>
<section>
<h1>model</h1>
<h3>Meshnet</h3>
</section>
<section>
<h2>the task</h2>
<img class="stretch" style="border:0; box-shadow: 0px 0px 0px
rgba(150, 150, 255, 0.8);" width="250"
src="figures/tasks.png" alt="david">
</section>
<section>
<h2>state of the art: freesurfer</h2>
<img class="stretch" style="border:0; box-shadow: 0px 0px 0px
rgba(150, 150, 255, 0.8);" width="250"
src="figures/freesurfer.png" alt="david">
<br>
<div class='slide-footer' style="text-align: left;">
Dale et al. Cortical surface-based analysis. I. Segmentation and
surface reconstruction. Neuroimage 1999
</div>
</section>
<section>
<h2>deep learning standard: U-net</h2>
<img class="stretch" style="border:0; box-shadow: 0px 0px 0px
rgba(150, 150, 255, 0.8);" width="250"
src="figures/unet_arc.png" alt="david">
<p>
<div class='slide-footer' style="text-align: left;">
Ronneberger et al. U-net: Convolutional networks
for biomedical image segmentation. MICCAI 2015<br>
Çiçek et al. 3D U-Net: learning dense volumetric
segmentation from sparse annotation. MICCAI 2016
</div>
</section>
<section>
<h2>deep learning standard: U-net</h2>
<img class="stretch" style="border:0; box-shadow: 0px 0px 0px
rgba(150, 150, 255, 0.8);" width="250"
src="figures/unet_table.png" alt="david">
<p>
<div class='slide-footer' style="text-align: left;">
Ronneberger et al. U-net: Convolutional networks for
biomedical image segmentation. MICCAI 2015<br>
Çiçek et al. 3D U-Net: learning dense volumetric
segmentation from sparse annotation. MICCAI 2016
</div>
</section>
<section>
<div id="header-right" style="right: -21%; z-index: 1500;">
<img width="200px" style="margin-bottom: -5%;"
src="figures/AlexFedorov.jpg" alt="Alex"><br>
<small>Alex Fedorov</small>
</div>
<h2>our model: Meshnet</h2>
<img src="figures/dilation.png" style="border:0; box-shadow: 0px
0px 0px rgba(150, 150, 255, 0.8);float: left;"
alt="dilated" width="40%">
<img src="figures/mn1.png" style="border:0; box-shadow: 0px
0px 0px rgba(150, 150, 255, 0.8);float: right;"
alt="Websocket" width="35%"
>
<div class="ulist">
<ul style="width: 55%; float: left; font-size:30px;">
<li>Gray and White matter</li>
<li>FreeSurfer for ground truth</li>
<li>T1 MRIs from HCP</li>
<li>GitHub <br>
<small>
<a href="https://github.com/Entodi/MeshNet">
https://github.com/Entodi/MeshNet</a>
</small>
</li>
</ul>
</div>
<br>
<div class='slide-footer' style="text-align: left;">
Fedorov et al. End-to-end learning of brain tissue segmentation
from imperfect labeling. IJCNN 2017
</div>
</section>
<section>
<h2>Meshnet</h2>
<img src="figures/mntable.png" style="border:0; box-shadow: 0px
0px
0px rgba(150, 150, 255,
0.8);float: right;"
alt="Websocket" width="60%" >
<div>
<ul style="width: 30%; float: left;
font-size:30px;">
<li>72516 vs. 23523355</li>
<li>600kb vs. 2Gb</li>
</ul>
</div>
<p>
<br>
<div class='slide-footer' style="text-align: left;">
Fedorov et al. End-to-end learning of brain tissue segmentation
from imperfect labeling. IJCNN 2017
</div>
</section>
<section>
<h2>Meshnet</h2>
<img src="figures/mnpipeline.png" style="border:0; box-shadow: 0px
0px 0px rgba(150, 150, 255, 0.8);"
alt="Websocket" class="stretch"
>
<br>
<div class='slide-footer' style="text-align: left;">
Fedorov et al. End-to-end learning of brain tissue segmentation
from imperfect labeling. IJCNN 2017
</div>
</section>
<section>
<h2>Meshnet</h2>
<img src="figures/mnexample.png" style="border:0; box-shadow: 0px
0px 0px rgba(150, 150, 255, 0.8);"
alt="Websocket" class="stretch"
>
<br>
<div class='slide-footer' style="text-align: left;">
Fedorov et al. End-to-end learning of brain tissue segmentation
from imperfect labeling. IJCNN 2017
</div>
</section>
<section>
<h3>better than the teacher</h3>
<img class="stretch" style="border:0; box-shadow: 0px 0px 0px
rgba(150, 150, 255, 0.8);" width="700"
src="figures/MN_examples.png" alt="loop">
</section>
<section>
<h3>better than the human (sometimes)</h3>
<img class="stretch" style="border:0; box-shadow: 0px 0px 0px
rgba(150, 150, 255, 0.8);" width="700"
src="figures/mnhuman.png" alt="loop">
</section>
<section>
<h3>better than U-net</h3>
<img class="stretch" style="border:0; box-shadow: 0px 0px 0px
rgba(150, 150, 255, 0.8);"
src="figures/mnvsunet.png" alt="loop">
</section>
<section>
<h3>an easy segmentation model</h3>
<div class="ulist">
<ul style="width: 60%; float:
center;">
<li>compact and portable</li>
<li>universal and adaptable</li>
<li>code:
<a href="https://github.com/Entodi/MeshNet">
https://github.com/Entodi/MeshNet</a>
</li>
</ul>
</div>
</section>
<section data-background-video="figures/brainchopV1_3.mp4">
</section>
</section>
<section>
<h2>Team that's working on this</h2>
<img class="stretch" style="border:0; box-shadow: 0px 0px 0px
rgba(150, 150, 255, 0.8);" width="800"
src="figures/the_team23.svg" alt="loop">
</section>
</div>
</div>
<script src="dist/reveal.js"></script>
<link rel="stylesheet" href="plugin/highlight/monokai.css">
<script src="plugin/highlight/highlight.js"></script>
<script src="plugin/math/math.js"></script>
<script src="plugin/chalkboard/plugin.js"></script>
<script src="plugin/notes/notes.js"></script>
<script src="plugin/zoom/zoom.js"></script>
<script src="plugin/fullscreen/fullscreen.js"></script>
<script src="plugin/menu/menu.js"></script>
<script src="plugin/verticator/verticator.js"></script>
<link rel="stylesheet" href="plugin/verticator/verticator.css">
<script>
// Full list of configuration options available at:
// https://github.com/hakimel/reveal.js#configuration
Reveal.initialize({
// history: true,
hash: true,
margin: 0.01,
minScale: 0.01,
maxScale: 1.23,
center: true,
controls: false,
keyboard: true,
overview: true,
transition: 'slide', // Transition style: none/fade/slide/convex/concave/zoom
transitionSpeed: 'slow', // Transition speed: default/fast/slow
menu: {
themes: false,
openSlideNumber: true,
openButton: false,
},
chalkboard: {
boardmarkerWidth: 1,
chalkWidth: 2,
chalkEffect: 1,
toggleNotesButton: false,
toggleChalkboardButton: false,
slideWidth: Reveal.width,
slideHeight: Reveal.height,
// src: "chalkboards/chalkboard_em2.json",
readOnly: false,
theme: "blackboard",
eraser: { src: "plugin/chalkboard/img/sponge.png", radius: 30},
},
math: {
mathjax: 'https://cdn.jsdelivr.net/gh/mathjax/[email protected]/MathJax.js',
config: 'TeX-AMS_SVG-full',
// pass other options into `MathJax.Hub.Config()`
TeX: {
Macros: {
RR: '\\mathbb{R}',
PP: '\\mathbb{P}',
EE: '\\mathbb{E}',
NN: '\\mathbb{N}',
vth: '\\vec{\\theta}',
loss: '{\\cal l}',
hclass: '{\\cal H}',
CD: '{\\cal D}',
def: '\\stackrel{\\text{def}}{=}',
pag: ['\\text{pa}_{{\cal G}^{#1}}(#2)}', 2],
vec: ['\\boldsymbol{\\mathbf #1}', 1],
set: [ '\\left\\{#1 \\; : \\; #2\\right\\}', 2 ],
bm: ['\\boldsymbol{\\mathbf #1}', 1],
argmin: ['\\operatorname\{arg\\,min\\,\}'],
argmax: ['\\operatorname\{arg\\,max\\,\}'],
prob: ["\\mbox{#1$\\left(#2\\right)$}", 2],
},
loader: {load: ['[tex]/color']},
extensions: ["color.js"],
tex: {packages: {'[+]': ['color']}},
svg: {
fontCache: 'global'
}
}
},
plugins: [ Verticator, RevealMath, RevealChalkboard, RevealHighlight, RevealNotes, RevealZoom, RevealMenu ],
});
Reveal.configure({ fragments: true }); // set false when developing to see everything at once
Reveal.configure({ slideNumber: true });
//Reveal.configure({ history: true });
Reveal.configure({ slideNumber: 'c / t' });
Reveal.addEventListener( 'darkside', function() {
document.getElementById('theme').setAttribute('href','dist/theme/aml_dark.css');
}, false );
Reveal.addEventListener( 'brightside', function() {
document.getElementById('theme').setAttribute('href','dist/theme/aml.css');
}, false );
</script>
<style type="text/css">
/* 1. Style header/footer <div> so they are positioned as desired. */
#header-left {
position: absolute;
top: 0%;
left: 0%;
}
#header-right {
position: absolute;
top: 0%;
right: 0%;
}
#footer-left {
position: absolute;
bottom: 0%;
left: 0%;
}
</style>
<!-- // 2. Create hidden header/footer -->
<div id="hidden" style="display:none;">
<div id="header">
<div id="header-left">
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1);" width="300"
src="figures/TReNDS_logo_light.png" alt="TReNDS logo">
</div>
<div id="header-right">
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1);" width="150"
src="figures/neuroneural.net.svg" alt="neuroneural.net">
</div>
<div id="footer-left">
</div>
</div>
</div>
<script type="text/javascript">
// 3. On Reveal.js ready event, copy header/footer <div> into each `.slide-background` <div>
var header = $('#header').html();
if ( window.location.search.match( /print-pdf/gi ) ) {
Reveal.addEventListener( 'ready', function( event ) {
$('.slide-background').append(header);
});
}
else {
$('div.reveal').append(header);
}
</script>
</body>
</html>