-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathDeepLearning.html
623 lines (566 loc) · 23.3 KB
/
DeepLearning.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
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
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>A practical introduction to Deep Learning</title>
<meta name="description" content="A practical introduction to Deep Learning">
<meta name="author" content="Giovanni Pellegrini">
<meta name="apple-mobile-web-app-capable" content="yes" />
<meta name="apple-mobile-web-app-status-bar-style" content="black-translucent" />
<meta name="viewport"
content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no, minimal-ui">
<link rel="stylesheet" href="./reveal.js/dist/reset.css">
<link rel="stylesheet" href="./reveal.js/dist/reveal.css">
<link rel="stylesheet" href="./reveal.js/dist/theme/black.css" id="theme">
<link rel="stylesheet" href="./reveal.js/plugin/highlight/zenburn.css">
<script src="./reveal.js/dist/reveal.js"></script>
<script src="./reveal.js/plugin/markdown/markdown.js"></script>
<script src="./reveal.js/plugin/math/math.js"></script>
<script src="./reveal.js/plugin/highlight/highlight.js"></script>
<!-- <script src="../reveal.js-plugins/menu/menu.js"></script>
<script src="../reveal.js-plugins/chalkboard/plugin.js"></script>
<script src="../reveal.js-plugins/audio-slideshow/plugin.js"></script> -->
<!-- Font awesome -->
<!-- <link rel="stylesheet" href="../reveal.js-plugins/menu/font-awesome/css/fontawesome.css"> -->
<!-- <script async defer src="https://buttons.github.io/buttons.js"></script> -->
<!-- custom css -->
<style>
.container {
display: flex;
}
.col {
flex: 1;
}
</style>
<!-- end custom css -->
</head>
<body>
<div class="reveal">
<!-- Any section element inside of this container is displayed as a slide -->
<div class="slides">
<!-- A practical introduction to deep learning -->
<section>
<h1 class="r-fit-text">A practical introduction to deep learning</h1>
<img style="height:800px" src="resources/nn.svg" alt="nn">
</section>
<!-- The rise of deep learning: why? -->
<section>
<h1 class="r-fit-text">The rise of deep learning: why?</h1>
<img style="height:800px" src="https://imgs.xkcd.com/comics/tasks.png" alt="xkcd">
</section>
<!-- Imagenet -->
<section data-background-image="resources/alexnet_paperswithcode.png">
</section>
<!-- The rise of deep learning: why? -->
<section>
<h1 class="r-fit-text">AlexNet: the mother of all networks</h1>
<div class="r-stack">
<img class="fragment fade-out" data-fragment-index="0"
src="https://paperswithcode.com/media/methods/Screen_Shot_2020-06-22_at_6.35.45_PM.png"
style="width: 1200px;">
<img class="fragment current-visible" data-fragment-index="0"
src="resources/alexnet.png"
style="width: 1000px;">
</div>
</section>
<!-- A simple formal definition of neural network -->
<section>
<h1 class="r-fit-text">A simple formal definition of neural network</h1>
<br><br>
<ul style="font-size: 2em; text-align:center; list-style-position: inside;">
<li>\( g_{_{W}}:\mathbb{R}^{n} \to \mathbb{R}^{m} \)</li>
<li>\( W \to \) neural network parameters (Weights)</li>
<li>\( g_{_{W}} \to \) differentiable everywhere</li>
</ul>
</section>
<!-- mnist: hand written digits classification -->
<section>
<h1 class="r-fit-text">Mnist: hand written digits</h1>
<h2>A classification problem</h2>
<div class="container" style="align-items:center;">
<div class="col">
<img style="height:500px" src="resources/MnistExamples.png" alt="mnist1">
</div>
<div class="col">
<p style="font-size: 1.5em;">\( g_{_{W}}:\mathbb{R}^{n} \times \mathbb{R}^{n} \to
[0,1]^{m} \)</p>
<p style="font-size: 1.5em;">\( n = 28 \, (pixels) \)</p>
<p style="font-size: 1.5em;">\( m = 10 \, (classes) \)</p>
</div>
</div>
</section>
<!-- Pixel by pixel processing -->
<section>
<h1 class="r-fit-text">Pixel by pixel processing</h1>
<h2>A semantic segmentation problem</h2>
<div class="container" style="align-items:center;">
<div class="col">
<img style="height:600px" src="resources/ThisSegmentationDoesNotExist.png" alt="segment1">
</div>
<div class="col">
<p style="font-size: 2em;">\( g_{_{W}}:\mathbb{R}^{n} \times \mathbb{R}^{n} \to
\mathbb{R}^{n} \times \mathbb{R}^{n} \)</p>
</div>
</div>
</section>
<!-- Fully connected neural networks -->
<section>
<h1 class="r-fit-text">Fully connected neural networks</h1>
<h2>aka multilayer perceptron</h2>
<div class="container" style="align-items:center;">
<div class="col">
<img style="height:600px" src="resources/nn.svg" alt="nn2">
</div>
<div class="col">
<h3 class="r-fit-text">Multilayer perceptron components</h3>
<ul style="font-size: 1.5em;">
<li>Input: \( x \in \mathbb{R}^{25} \)</li>
<li>Output: \( y \in \mathbb{R}^{10} \)</li>
<li>Weights: \( W = (w_{jk}) \)</li>
<li>Activation functions: \( f \)</li>
</ul>
</div>
</div>
</section>
<!-- Forward propagation: weight multiplication -->
<section>
<h1 class="r-fit-text">Fully connected neural networks: forward propagation</h1>
<div class="container" style="align-items:center;">
<div class="col">
<img style="height:600px" src="resources/nn.svg" alt="nn2">
</div>
<div class="col">
<p style="font-size: 1.5em;">\( z_{1} = W^{1} x \)</p>
<p style="font-size: 1.5em;">\( a_{1} = f^{1}(z_{1}) \)</p>
<p style="font-size: 1.5em;">\( \Downarrow \)</p>
<p style="font-size: 1.5em;">\( z_{l} = W^{l} a_{l-1} \)</p>
<p style="font-size: 1.5em;">\( a_{l} = f^{l}(z_{l}) \)</p>
<p style="font-size: 1.5em;">\( \Downarrow \)</p>
<p style="font-size: 1.5em;">\( y = f^{L}(z_{L}) \)</p>
</div>
</div>
</section>
<!-- Activation functions -->
<section>
<h1>Activation functions</h1>
<div class="container" style="align-items:center;">
<div class="col">
<div class="r-stack">
<img class="fragment fade-out" data-fragment-index="0"
src="resources/Activation_logistic.svg" style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="0"
src="resources/Activation_tanh.svg" style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="1"
src="resources/Activation_rectified_linear.svg" style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="2"
src="resources/Activation_prelu.svg" style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="3"
src="resources/Activation_softplus.svg" style="height: 500px;">
</div>
</div>
<div class="col">
<div class="r-stack">
<p class="fragment fade-out" data-fragment-index="0" style="font-size: 2em;">Sigmoid \[
f(x)=\frac{1}{1 + e^{-x}} \]</p>
<p class="fragment current-visible" data-fragment-index="0" style="font-size: 2em;">Tanh \[
f(x)=\frac{e^{x}-e^{-x}}{e^{x} + e^{-x}} \]</p>
<p class="fragment current-visible" data-fragment-index="1" style="font-size: 2em;">ReLU\[
f(x)=max\{0,x\} \]</p>
<p class="fragment current-visible" data-fragment-index="2" style="font-size: 2em;">Leaky
ReLU\[ f(x)=f(x)=max\{0.01x,x\} \]</p>
<p class="fragment current-visible" data-fragment-index="3" style="font-size: 2em;">Softplus
\[ f(x)=ln(1 + e^{x}) \]</p>
</div>
</div>
</div>
</section>
<!-- Problems of a fully connected neural network -->
<section>
<h1 class="r-fit-text">Problems of a fully connected neural network</h1>
<div class="container" style="align-items:center;">
<div class="col">
<img style="height:600px" src="resources/nn.svg" alt="nn2">
</div>
<div class="col">
<ul style="font-size: 1.5em;">
<li>Does not respect positional correlation for an image</li>
<li>It is not sparse</li>
<li>Scales poorly for large inputs</li>
</ul>
</div>
</div>
</section>
<!-- Convolutional neural networks -->
<section>
<h1>Convolutional neural networks</h1>
<img style="height:600px" src="resources/cnn.svg" alt="cnn">
</section>
<!-- Examples of convolution layers -->
<section>
<h1>Examples of convolutions</h1>
<div class="container" style="align-items:center;">
<div class="col">
<div class="r-stack">
<img class="fragment fade-out" data-fragment-index="0"
src="resources/no_padding_no_strides.gif" style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="0"
src="resources/same_padding_no_strides.gif" style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="1"
src="resources/Convolutional_Neural_Network_NeuralNetworkFeatureLayers.gif"
style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="2"
src="resources/3_filters_in_a_Convolutional_Neural_Network.gif" style="height: 500px;">
</div>
</div>
<div class="col">
<blockquote>
<ul style="text-align:center; list-style-position: inside; font-size: 0.8em;">
<li>A convolution window (also called filter) moves across the input and produces and output with element by element multiplication</li>
<br>
<li>The weights \( W \) are the parameters of the filter (9 in this case). The convolution is defined by a size (3x3) and a stride (1x1)</li>
<br>
<li>Multiple filters can be used over a single input</li>
<br>
<li>The input can be padded in different ways and for different reasons (for instance to
preserve the input dimensions)</li>
</ul>
</blockquote>
</div>
</div>
</section>
<!-- An example of pooling layers: max-pooling-->
<section>
<h1 class="r-fit-text">An example of pooling layers: max-pooling</h1>
<img style="height:800px" src="resources/poolfig.gif" alt="maxpool">
<p>Credits to: <a href="https://mlnotebook.github.io/">https://mlnotebook.github.io/</a></p>
</section>
<!-- Convolutional neural networks -->
<section>
<h1>Convolutional neural networks</h1>
<img style="height:600px" src="resources/cnn.svg" alt="cnn">
</section>
<!-- Convnet demo-->
<section>
<h1 class="r-fit-text">An interactive live demo of a ConvNet</h1>
<a href="https://www.cs.ryerson.ca/~aharley/vis/conv/">https://www.cs.ryerson.ca/~aharley/vis/conv/</a>
</section>
<!-- Introduzione alla temperatura -->
<section>
<h1>The machine learning paradigm shift</h1>
<img style="height:500px" src="resources/machine_learning.jpg" alt="maxpool">
</section>
<!-- Supervised learning: mnist -->
<section>
<h1>Supervised learning: mnist</h1>
<div class="container" style="align-items:center;">
<div class="col">
<img style="height:500px" src="resources/mnist_label.png" alt="mnist_label">
</div>
<div class="col">
<blockquote>
The dataset consists of input data (images) and output data (labels). When trained on a
large amount of hand labeled data the network can learn to correctly classify previously
unseen images. A well trained neural network outputs correct labels.
</blockquote>
</div>
</div>
</section>
<!-- Loss function -->
<section>
<h1>Loss function</h1>
<div class="container" style="align-items:center;">
<div class="col">
<p style="font-size: 2.5em;"> \[ Loss = \mathcal{L}(g_{_{W}}(x),y) \]</p>
</div>
<div class="col" style="font-size: 1.25em;">
<p>\[\mathcal{L}(g_{_{W}}(x),y) = \sum _{i}^{data} \lvert \lvert g_{_{W}}(x_{i}) - y_{i} \rvert
\rvert_{1} \]</p>
<p>\[\mathcal{L}(g_{_{W}}(x),y) = \sum _{i}^{data} \lvert \lvert g_{_{W}}(x_{i}) - y_{i} \rvert
\rvert_{2} \]</p>
<p>\[\mathcal{L}(g_{_{W}}(x),y) = \sum _{i}^{data} CE(g_{_{W}}(x_{i}) - y_{i}) \]</p>
</div>
</div>
</section>
<!-- Minimizing the loss function -->
<section>
<h1>Minimizing the loss function</h1>
<h2>Gradient descent</h2>
<div class="container" style="align-items:center;">
<div class="col">
<img src="resources/Gradient_descent.gif" alt="pressure" , style="width: 400px;">
</div>
<div class="col" style="font-size: 1.25em;">
<ol>
<li>\( W_{0}, \mathcal{L}(g_{_{W_{0}}}(x),y)\)</li>
<br>
<li>\( \nabla \mathcal{L}(g_{_{W_{0}}}(x),y)\)</li>
<br>
<li>\( W_{1} = W_{0} - \gamma \nabla \mathcal{L}(g_{_{W_{0}}}(x),y) \)</li>
<p>...</p>
<li>\( W_{n+1} = W_{n} - \gamma \nabla \mathcal{L}(g_{_{W_{n}}}(x),y) \)</li>
<p>\( \gamma = learning \, rate \)</p>
</ol>
<p> </p>
</div>
</div>
</section>
<!-- Minimizing the loss function -->
<section>
<h1 class="r-fit-text">Minimizing faster: batch gradient descent</h1>
<div class="container" style="align-items:center;">
<div class="col">
<p style="font-size: 1.5em;">\[ \mathcal{L}_{batch} = \sum _{i}^{n_{batch}}
(g_{_{W_{0}}}(x_{i}),y_{i})\]</p>
<br>
<p style="font-size: 1.5em;">\[ n_{batch} = 32 \, (64, 128) \]</p>
</div>
<div class="col" style="font-size: 1.25em;">
<ol>
<li>\( W_{0}, \mathcal{L}_{batch}(g_{_{W_{0}}}(x),y)\)</li>
<br>
<li>\( \nabla \mathcal{L}(_{batch}g_{_{W_{0}}}(x),y)\)</li>
<br>
<li>\( W_{1} = W_{0} - \gamma \nabla \mathcal{L}_{batch}(g_{_{W_{0}}}(x),y) \)</li>
<p>...</p>
<li>\( W_{n+1} = W_{n} - \gamma \nabla \mathcal{L}_{batch}(g_{_{W_{n}}}(x),y) \)</li>
<p>\( \gamma = learning \, rate \)</p>
</ol>
<p> </p>
</div>
</div>
</section>
<!-- How to compute the gradient? -->
<section>
<h1>How to compute the gradient?</h1>
<h2>Backpropagation, or the chain rule</h2>
<img style="height: 500px;" src="resources/backpropagation.jpeg" alt="backprop">
</section>
<!-- Training a network in practice -->
<section>
<h1>Training a network in practice</h1>
<div class="container" style="align-items:center;">
<div class="col">
<div class="r-stack">
<img class="fragment fade-out" data-fragment-index="0" src="resources/cnn.svg"
style="width: 900px;">
<img class="fragment current-visible" data-fragment-index="0"
src="resources/Gradient_descent.gif" style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="1"
src="resources/MnistExamplesTraining.png" style="height: 500px;">
<img class="fragment current-visible" data-fragment-index="2"
src="resources/MnistExamplesTest.png" style="height: 500px;">
</div>
</div>
<div class="col">
<div class="r-stack">
<p class="fragment fade-out" data-fragment-index="0" style="font-size: 2em;"> The Network
</p>
<p class="fragment current-visible" data-fragment-index="0" style="font-size: 2em;">\[
\mathcal{L}(g_{_{W}}(x),y) \]</p>
<p class="fragment current-visible" data-fragment-index="1" style="font-size: 2em;">The
Data: Training Set</p>
<p class="fragment current-visible" data-fragment-index="2" style="font-size: 2em;">The
Data: Validation Set</p>
</div>
</div>
</div>
</section>
<!-- Is It working? -->
<section>
<h1>Is It working?</h1>
<img style="height: 500px;" src="resources/loss_accuracy.png" alt="backprop">
<p>Epoch \( \Rightarrow \) One full pass through the whole training set</p>
</section>
<!-- Deep learning for Raman Spectroscopy -->
<section>
<h1 class="r-fit-text">Deep learning for Raman Spectroscopy</h1>
<h2>The state of the art</h2>
<div class="col">
<div class="r-stack">
<img class="fragment fade-out" data-fragment-index="0" src="resources/raman_lit1.png"
style="width: 900px;">
<img class="fragment current-visible" data-fragment-index="0" src="resources/raman_lit2.png"
style="width: 900px;">
<img class="fragment current-visible" data-fragment-index="1" src="resources/raman_lit3.png"
style="width: 900px;">
<img class="fragment current-visible" data-fragment-index="2" src="resources/raman_lit4.png"
style="width: 900px;">
</div>
</div>
</section>
<!-- Deep learning for Raman Spectroscopy: the data -->
<section>
<h1 class="r-fit-text">Deep learning for Raman Spectroscopy: the data</h1>
<div class="container" style="align-items:center;">
<div class="col">
<img style="height: 900px;" src="resources/dataset.png" alt="backprop">
</div>
<div class="col" style="font-size: 1.25em;">
<blockquote>
The RUFF dataset consists of more than 1600 different classes of spectra, with an uneven
number of spectra for each class
</blockquote>
<br>
<blockquote>
The dataset was split in training and validation set leaving just one sample spectrum per
class in the validation set
</blockquote>
</div>
</div>
</section>
<!-- Deep learning for Raman Spectroscopy: the network -->
<section>
<h1 class="r-fit-text">Deep learning for Raman Spectroscopy: the network</h1>
<div class="container" style="align-items:center;">
<div class="col" , style="width: 1500px;">
<img style="width: 1200px;" src="resources/raman_cnn.png" alt="raman_cnn">
</div>
<div class="col" style="font-size: 1.25em;">
<blockquote>
The network is a one-dimensional convolutional neural network that takes a Raman spectrum as
an input and outputs a chemical compound class as an output
</blockquote>
</div>
</div>
</section>
<!-- Deep learning for Raman Spectroscopy: the performance -->
<section>
<h1 class="r-fit-text">Deep learning for Raman Spectroscopy: the performance</h1>
<img style="width: 1700px;" src="resources/raman_perf1.png" alt="backprop">
<br>
<img style="width: 17 00px;" src="resources/raman_perf2.png" alt="backprop">
</section>
<!-- Our approach to Raman and Deep Learning: the data -->
<section>
<h1 class="r-fit-text">Our approach to Raman and Deep Learning: the data</h1>
<div class="container" style="align-items:center;">
<div class="col">
<img style="width: 1000px;" src="resources/raman_lit3.png" alt="backprop">
</div>
<div class="col" style="font-size: 1.25em;">
<img style="width: 800px;" src="resources/raman_mydata.png" alt="backprop">
<blockquote>
Our dataset consists of 1322 different classes of preprocessed spectra, with 12 to 20
spectral samples for each class. One third of the data is assigned to the validation
dataset.
</blockquote>
<br>
</div>
</div>
</section>
<!-- Data augmentation -->
<section>
<h1>Data augmentation</h1>
<div class="container" style="align-items:center;">
<div class="col">
<div class="r-stack">
<img class="fragment fade-out" data-fragment-index="0" src="resources/data_augment_none.png"
style="width: 1200px;">
<img class="fragment current-visible" data-fragment-index="0"
src="resources/data_augment_noise.png" style="width: 1200px;">
<img class="fragment current-visible" data-fragment-index="1"
src="resources/data_augment_baseline.png" style="width: 1200px;">
<img class="fragment current-visible" data-fragment-index="2"
src="resources/data_augment_magnitude.png" style="width: 1200px;">
<img class="fragment current-visible" data-fragment-index="3"
src="resources/data_augment_energy.png" style="width: 1200px;">
<img class="fragment current-visible" data-fragment-index="4"
src="resources/data_augment_all.png" style="width: 1200px;">
</div>
</div>
<div class="col">
<div class="r-stack">
<p class="fragment fade-out" data-fragment-index="0" style="font-size: 2.5em;"> Original
</p>
<p class="fragment current-visible" data-fragment-index="0" style="font-size: 3.5em;">Noise
</p>
<p class="fragment current-visible" data-fragment-index="1" style="font-size: 3.5em;">
Baseline</p>
<p class="fragment current-visible" data-fragment-index="2" style="font-size: 3.5em;">
Magnitude Warping</p>
<p class="fragment current-visible" data-fragment-index="3" style="font-size: 3.5em;">Energy
Warping</p>
<p class="fragment current-visible" data-fragment-index="4" style="font-size: 3.5em;">All
Augmentations</p>
</div>
</div>
</div>
</section>
<!-- Our approach to Raman and Deep Learning: the network -->
<section>
<h1 class="r-fit-text">Our approach to Raman and Deep Learning: the network</h1>
<img style="width: 1500px;" src="resources/raman_mycnn.png" alt="backprop">
</section>
<!-- Our approach to Raman and Deep Learning: the results -->
<section>
<h1 class="r-fit-text">Our approach to Raman and Deep Learning: the results</h1>
<blockquote style="font-size: 4em;">
96% accuracy on the validation set
</blockquote>
</section>
<!-- Transitioning to our data: transfer learning -->
<section>
<h1 class="r-fit-text">Transitioning to our data: transfer learning</h1>
<div class="r-stack">
<img class="fragment fade-out" data-fragment-index="0" src="resources/transfer_learning_pre.png"
style="width: 1200px;">
<img class="fragment current-visible" data-fragment-index="0" src="resources/transfer_learning.png"
style="width: 1200px;">
<blockquote class="fragment current-visible" data-fragment-index="1" style="font-size: 3em;">
Retraining is always boring, and requires a GPU!!!
</blockquote>
</div>
<a
href="https://medium.com/@albertoarrigoni/transfer-learning-35fc47f081b5">https://medium.com/@albertoarrigoni/transfer-learning-35fc47f081b5</a>
</section>
<!-- A completely different approach: self-supervised learning -->
<section>
<h1>A completely different approach</h1>
<h2>Self-supervised learning</h2>
<div class="r-stack">
<img class="fragment fade-out" data-fragment-index="0" src="resources/barlow.png"
style="width: 1200px;">
<img class="fragment current-visible" data-fragment-index="0" src="resources/mnist_tsne.png"
style="width: 900px;">
<blockquote class="fragment current-visible" data-fragment-index="1" style="font-size: 3em;">
No retraining and potentially no need for supervision!!!
</blockquote>
</div>
</section>
<!-- Let's try it ourselves -->
<section>
<h1>Let's try it ourselves</h1>
<a href="https://colab.research.google.com/github/AviatorMoser/keras-mnist-tutorial/blob/master/MNIST%20in%20Keras.ipynb">Python + Keras + Colaboratory</a>
</section>
<!-- Wrapping up -->
<section>
<h1>Wrapping up</h1>
<img style="height:800px" src="https://imgs.xkcd.com/comics/machine_learning.png" alt="xkcd">
</section>
</div>
</div>
<script>
// Full list of configuration options available here:
// https://github.com/hakimel/reveal.js#configuration
Reveal.initialize({
slideNumber: true,
controls: true,
progress: true,
history: true,
center: true,
mouseWheel: true,
navigationMode: 'linear',
width: 1920,
height: 1080,
minScale: 0.1,
maxScale: 4.0,
plugins: [RevealMarkdown, RevealHighlight, RevealMath],
});
function changeTheme(input) {
var config = {};
config.theme = input.value;
input.blur();
}
</script>
</body>
</html>