-
Notifications
You must be signed in to change notification settings - Fork 0
/
verona_ad.html
636 lines (565 loc) · 33.8 KB
/
verona_ad.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
624
625
626
627
628
629
630
631
632
633
634
635
636
<!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 rel="stylesheet" href="https://maxcdn.bootstrapcdn.com/font-awesome/4.5.0/css/font-awesome.min.css">
<!-- <link href="https://stackpath.bootstrapcdn.com/font-awesome/4.7.0/css/font-awesome.min.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>(Extremely) Quick DL Intro</title>
<meta name="description" content="Lectures at University of Verona">
<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>
<h1>Automatic Differentiation</h1>
<h2>Selected topics in Deep Learning</h2>
<h3>in 6 hours or less</h3>
<h3>by Sergey Plis</h3>
<p>
</section>
</section>
<section>
<section>
<h2>Algorithmic Differentiation (AD)</h2>
<blockquote style="width: 100%;">
".. may be one of the best scientific computing techniques you’ve never heard of."
<br>
<a href="https://alexey.radul.name/ideas/2013/introduction-to-automatic-differentiation/">Alexey Radul</a>
</blockquote>
</section>
<section>
<blockquote style="background-color: #eee8d5; width: 100%;" class="fragment" data-fragment-index="0">
The very first computer science PhD dissertation introduced forward accumulation mode automatic differentiation.
</blockquote>
<blockquote style="background-color: #93a1a1; color: #fdf6e3; font-size: 38px;" class="fragment" data-fragment-index="1">
<center>
Wengert (1964)
</center>
</blockquote>
</section>
<section data-background="figures/AD_paper.png" background-size="cover">
<blockquote style="background-color: #93a1a1; color: #fdf6e3; width: 100%">
Robert Edwin Wengert. A simple automatic derivative evaluation program. Communications of the ACM 7(8):463–4, Aug 1964.
</blockquote>
<blockquote style="background-color: #eee8d5; width: 100%; font-size: 32px;" class="left bordered">
A procedure for automatic evaluation of total/partial derivatives of arbitrary algebraic functions is presented. The technique permits computation of numerical values of derivatives without developing analytical expressions for the derivatives. <span class="fragment highlight-red" data-fragment-index="0">The key to the method is the decomposition of the given function</span>, by introduction of intermediate variables, <span class="fragment highlight-red" data-fragment-index="0">into a series of elementary functional steps</span>. A library of elementary function subroutines is provided for the automatic evaluation and differentiation of these new variables. The final step in this process produces the desired function’s derivative. The main feature of this approach is its simplicity. It can be used as a quick-reaction tool where the derivation of analytical derivatives is laborious and also as a debugging tool for programs which contain derivatives.
</blockquote>
</section>
<section data-background="figures/wengert.png" background-size="cover">
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1);" width="1000"
src="figures/wengert2.png" class="bordered" alt="Wengert dissertation">
</section>
<section>
<blockquote style="background-color: #93a1a1; color: #fdf6e3; width: 100%">
R. E. Bellman, H. Kagiwada, and R. E. Kalaba (1965) Wengert’s numerical method for partial derivatives, orbit determination and quasilinearization, Communications of the ACM 8(4):231–2, April 1965, doi:10.1145/363831.364886
</blockquote>
<blockquote style="background-color: #eee8d5; font-size: 34px; width: 100%" class="left">
In a recent article in the Communications of the ACM, R. Wengert suggested a technique for machine evaluation of the partial derivatives of a function given in analytical form. In solving nonlinear boundary-value problems using quasilinearization many partial derivatives must be formed analytically and then evaluated numerically. <span class="fragment highlight-red" data-fragment-index="0">Wengert’s method</span> appears very attractive from the programming viewpoint and <span class="fragment highlight-red" data-fragment-index="0">permits the treatment of large systems</span> of differential equations <span class="fragment highlight-red" data-fragment-index="0">which might not otherwise be undertaken</span>.
</blockquote>
</section>
<section>
<blockquote style="background-color: #eee8d5; font-size: 40px;" class="left">
Automatic Differentiation (AD) mechanically calculates the derivatives (Leibnitz, 1664; Newton, 1704) of functions expressed as computer programs, at machine precision, and with complexity guarantees.
</blockquote>
</section>
<section>
<h2>Automatic Differentiation</h2>
<ul style="list-style-type: disk;">
<li class="fragment roll-in" data-fragment-index="1"> Derivative of $f: \RR^n \to \RR^m$ is $m\times n$ "Jacobian matrix" $\bm{J}$.
<li class="fragment roll-in" data-fragment-index="2"> AD in the forward accumulation mode: $\bm{J}v$ (Wengert, 1964)
<li class="fragment roll-in" data-fragment-index="3"> AD in the reverse accumulation mode, $\bm{J}^Tv$ (Speelpenning, 1980)
<li class="fragment roll-in" data-fragment-index="4"> About a zillion other modes and tricks
<li class="fragment roll-in" data-fragment-index="5"> Vibrant field with regular workshops, conferences, updated community portal
(<a href="http://autodiff.org" target="_blank">http://autodiff.org</a>)
</ul>
</section>
<section>
<h2>What is AD?</h2>
<ul style="list-style-type: none; font-size: 22pt">
<li class="fragment roll-in" data-fragment-index="0"> <em>Automatic Differentiation</em>
<ul style="list-style-type: none; font-size: 22pt">
<li class="fragment roll-in" data-fragment-index="1"> aka <em>Algorithmic Differentiation</em>
<ul style="list-style-type: none; font-size: 22pt">
<li class="fragment roll-in" data-fragment-index="2"> aka <em>Computational Differentiation</em>
</ul>
</ul>
</ul>
<dl style="list-style-type: disk; font-size: 22pt">
<span class="fragment roll-in" data-fragment-index="3">
<dt>AD Type I</dt>
<dd>A calculus for efficiently calculating derivatives of functions specified by a set of equations.
</dd>
</span>
<span class="fragment roll-in" data-fragment-index="4">
<dt>AD Type II</dt>
<dd>A way of transforming a computer program implementing a numeric function to also efficiently calculate some derivatives.</dd>
</span>
<span class="fragment roll-in" data-fragment-index="5">
<dt>AD Type III</dt>
<dd>
A computer program which automatically transforms an input computer program specifying a numeric function into one that also efficiently calculates derivatives.
</dd>
</span>
</dl>
</section>
<section data-background="figures/logisticbifurcation.png" background-size="cover">
<h2 style="text-shadow: 4px 4px 4px #002b36; color: #93a1a1">
AD is not Numerical Differentiation</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="1000"
src="figures/ADnotND.svg" alt="ADnotND">
<blockquote style="background-color: #eee8d5; font-size: 40px; width: 100%;">
Numerical diffirentiation is a problematic approximation
<ul style="list-style-type: disk; font-size: 22pt">
<li> though shalt not add small numbers to big numbers
<li> though shalt not subtract numbers which are approximately equal
</ul>
</blockquote>
</section>
<section data-background="figures/logisticbifurcation.png" background-size="cover">
<h3 style="text-shadow: 4px 4px 4px #002b36; color: #93a1a1">
What is wrong with Numerical Differentiation
</h3>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="80%"
src="figures/truncation_error.svg" alt="round off and truncation errors">
<blockquote style="background-color: #eee8d5; font-size: 40px;">
Numerical diffirentiation is a problematic approximation
<ul style="list-style-type: disk; font-size: 22pt">
<li> though shalt not add small numbers to big numbers
<li> though shalt not subtract numbers which are approximately equal
</ul>
</blockquote>
</section>
<section data-background="figures/logisticbifurcation.png" background-size="cover">
<h2 style="text-shadow: 4px 4px 4px #002b36; color: #93a1a1">
AD is not Symbolic Differentiation
</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0); margin-bottom: -10px; margin-top: -10px;" width="1000"
src="figures/ADsnotSymbolic.svg" alt="ADnotSymbolic">
<blockquote style="background-color: #eee8d5; font-size: 22px; width: 100%;" style="margin-top: -40px;">
\begin{align}
f(\vec{x}) & = \prod_{i=1}^d x_i \\
\nabla f(\vec{x}) & = \left(\frac{\partial f}{\partial x_1}, \frac{\partial f}{\partial x_2}, \dots, \frac{\partial f}{\partial x_d},\right)
= \left(\begin{array}{ccccc}
x_2x_3x_4\cdots x_{d-1}x_d, \\
x_1x_3x_4\cdots x_{d-1}x_d, \\
x_1x_2x_4\cdots x_{d-1}x_d, \\
\vdots \\
x_1x_2x_3\cdots x_{d-1} \\
\end{array}\right)
\end{align}
</blockquote>
<div class="slide-footer">
Example from Speelpenning, figure modified from Baydin
</div>
<div class="slide-footer">
For an alternative opinion see: <a href="https://arxiv.org/abs/1904.02990v3" target="blank_">On the Equivalence of Forward Mode Automatic Differentiation and Symbolic Differentiation
</a>
</div>
</section>
<section>
<h2>
We want exact derivative at a point!
</h2>
<img width="100%" src="figures/example_derivatives.svg" alt="Derivatives">
</section>
</section>
<!-- ------------------------------------------------------------------------- -->
<section>
<section>
<h1>Forward Mode AD</h1>
</section>
<section data-background="figures/logisticbifurcation.png" background-size="cover">
<h2 style="text-shadow: 4px 4px 4px #002b36; color: #93a1a1">
AD is</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="1000"
src="figures/ADis.svg" alt="ADis">
<div class="slide-footer">
Figure modified from Baydin
</div>
</section>
<section data-background="figures/logisticbifurcation.png" background-size="cover">
<h2>AD is</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0); margin-bottom: -10px; margin-top: -70px;" width="780"
src="figures/whatisADnot.svg" alt="AD">
<div class="slide-footer">
Figure from Baydin
</div>
</section>
<section>
<h2>AD graph</h2>
$$
y = f(x_1, x_2) = \ln(x_1) + x_1x_2 - \sin(x_2)
$$
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="800"
src="figures/computation_graph.svg" alt="graph">
</section>
<section>
<div id="header-left" style="font-size: 24px; left: -15%; top: -3%">
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1); " width="500"
src="figures/computation_graph.svg" alt="dice">
</div>
<h2>AD trace</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="75%"
src="figures/computation_trace.svg" alt="trace">
</section>
<section data-fullscreen>
<div id="header-left" style="font-size: 24px; left: -1%; top: -3%">
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1); " width="500"
src="figures/computation_graph.svg" alt="dice">
</div>
<h2>AD</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="130%"
src="figures/forwardAD_example.svg" alt="forward">
</section>
<section>
<h2>AD: a Jacobian column in a pass</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="1000"
src="figures/JacobianAD.svg" alt="forward">
</section>
<section>
<h2>AD: directional derivative</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="1000"
src="figures/JacobianADdirectional.svg" alt="directional">
</section>
<section>
<h3>Dual numbers (1873)</h3>
\begin{align}
v + \dot{v}\epsilon &\\
v, \dot{v} \in \RR, & \epsilon \ne 0, \epsilon^2 = 0 \\
(v + \dot{v}\epsilon) + (u + \dot{u}\epsilon) & = (v + u) + (\dot{v} + \dot{u})\epsilon\\
(v + \dot{v}\epsilon)(u + \dot{u}\epsilon) & = (vu) + (v\dot{u} + u\dot{v})\epsilon\\
f(v+\dot{v}\epsilon) & = f(v) + f^{\prime}(v)\dot{v}\epsilon
\end{align}
<div class="slide-footer">
Clifford (1873)
</div>
</section>
</section>
<!-- ------------------------------------------------------------------------- -->
<section>
<section>
<h1>Back propagation</h1>
</section>
<section>
<h2>the classical presentation</h2>
<img style="border:0; box-shadow: 0px 0px 0px rgba(255, 255, 255, 0);" width="780"
src="figures/backprop.svg" alt="BP">
<div class="slide-footer">
Figure from Baydin
</div>
</section>
</section>
<!-- ------------------------------------------------------------------------- -->
<section>
<section>
<h1>Reverse Mode AD</h1>
</section>
<section>
<blockquote style="background-color: #eee8d5; width: 100%;" class="left">
In the 1970s, tools for automated generation of adjoint codes (aka reverse accumulation mode automatic differentiation, aka backpropagation) were developed.
</blockquote>
<dl style="list-style-type: disk; font-size: 22pt">
<dt>Type I</dt>
<dd> Geniuses transforming mathematical systems
(Gauss; Feynman (1939); Rozonoer and Pontryagin (1959))
</dd>
<dt>Type II</dt>
<dd> Manual transformation of computational processes
(Bryson (1962); Werbos (1974); Le Cun (1985); Rumelhart et al. (1986))
</dd>
<dt>Type III</dt>
<dd> Computer programs transform other computer programs
(Speelpenning (1980); LUSH; TAPENADE)
</dd>
<dt>Type IV</dt>
<dd> First-class AD operators; closure
(STALIN$\nabla$; R$^6$RS-AD; AUTOGRAD; DIFFSHARP)
</dd>
</dl>
</section>
<section>
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1); " width="400"
src="figures/Speelpenning.jpeg" alt="Speelpenning">
<blockquote style="background-color: #93a1a1; color: #fdf6e3;">
<center>
Bert Speelpenning
</center>
</blockquote>
</section>
<section data-background="figures/SPAD_1.png" background-size="cover">
</section>
<section data-background="figures/SPAD_2.png" background-size="cover">
</section>
<section data-background="figures/SPAD_3.png" background-size="cover">
</section>
<section data-background="figures/SPAD_4.png" background-size="cover">
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1); " width="1200"
src="figures/SPAD_4w.png" alt="Speelpenning">
</section>
<section>
<blockquote style="background-color: #eee8d5; width: 100%; font-size: 22pt" class="fragment" data-fragment-index="0">
\begin{align}
y &= \log(\sin(x^2))
\end{align}
</blockquote>
<row>
<col80>
<ul style="list-style-type: none; font-size: 22pt">
<li class="fragment roll-in" data-fragment-index="1"> Traces:
<li class="fragment roll-in" data-fragment-index="2"> Primal
<li class="fragment roll-in" data-fragment-index="3"> Tangent Derivative
<li class="fragment roll-in" data-fragment-index="4"> Cotangent Derivative
</ul>
</col80>
<col80>
<ul style="list-style-type: none; font-size: 22pt">
<li class="fragment roll-in" data-fragment-index="1"> Direction:
<li class="fragment roll-in" data-fragment-index="2"> $\rightarrow$ Forward
<li class="fragment roll-in" data-fragment-index="3"> $\rightarrow$ Forward
<li class="fragment roll-in" data-fragment-index="4"> $\leftarrow$ Reverse
</ul>
</col80>
</row>
</section>
<section>
<h3>Computation Graph</h3>
<blockquote style="background-color: #93a1a1; color: #fdf6e3; font-size: 38px;" >
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1);" width="1000"
src="figures/graph.png" alt="computational graph">
</blockquote>
<ul style="list-style-type: none; font-size: 22pt">
<li class="fragment roll-in" data-fragment-index="0"> Following precedence rules
<li class="fragment roll-in" data-fragment-index="1"> binary/n-ary operators allowed $\rightarrow$ DAG (tree)
<li class="fragment roll-in" data-fragment-index="1"> $y =$ root, $x =$ leaves
</ul>
</section>
<section>
<h3>Intermediate Variables: $z_i$</h3>
<blockquote style="background-color: #eee8d5; width: 100%; font-size: 22pt" class="fragment" data-fragment-index="0">
\begin{align}
x &\\
z_1 &= x^2\\
z_2 &= \sin(z_1)\\
z_3 &= \log(z_2)\\
y &= z_3\\
\end{align}
</blockquote>
</section>
<section>
<h3>Adjoint: $\bar{z_i} = \frac{\partial y}{\partial z_i}$</h3>
</section>
<section >
<h3>Example</h3>
<blockquote style="background-color: #eee8d5; width: 100%; font-size: 22px" class="fragment" data-fragment-index="0">
<div class="row">
<div class="col-40">
\begin{align}
x &\\
z_1 &= x^2\\
z_2 &= \sin(z_1)\\
z_3 &= \log(z_2)\\
y &= z_3\\
\end{align}
</div>
<div class="col">
<ul style="font-size: 26px;">
<li class="fragment roll-in" data-fragment-index="5"> $\bar{x} = \left(\left(\frac{\partial y}{\partial z_3}\frac{\partial z_3}{\partial z_2}\right) \frac{\partial z_2}{\partial
l z_1}\right)\frac{\partial z_1}{\partial x}$
<li class="fragment roll-in" data-fragment-index="4"> $\bar{x} = \frac{\partial y}{\partial x} = \frac{\partial y}{\partial z_1}\frac{\partial z_1}{\partial x} = \bar{z}_1 2 x$
<li class="fragment roll-in" data-fragment-index="3"> $\bar{z}_1 = \frac{\partial y}{\partial z_1} = \frac{\partial y}{\partial z_2}\frac{\partial z_2}{\partial z_1} = \bar{z}_
2\mathrm{cos}(z_1)$
<li class="fragment roll-in" data-fragment-index="2"> $\bar{z}_2 = \frac{\partial y}{\partial z_2} = \frac{\partial y}{\partial z_3}\frac{\partial z_3}{\partial z_2} = \bar{y}\
\frac{1}{z_2}$
<li class="fragment roll-in" data-fragment-index="1"> $\bar{z}_3 = \frac{\partial y}{\partial z_3} = \frac{\partial y}{\partial y} = \bar{y} = 1$ (seed)
</ul>
</div>
</div>
</blockquote>
<blockquote style="background-color: rgba(0,0,0,0); width: 100%; font-size: 20px" class="fragment" data-fragment-index="6">
\begin{align}
\bar{x} & \fragment{6}{= \bar{z}_1 2 x}\\
&\fragment{7}{= (\bar{z}_2\mathrm{cos}(z_1) ) 2 x}\\
&\fragment{8}{= \left(\left(\bar{z}_3 \frac{1}{z_2} \right) \mathrm{cos}(x^2) \right) 2 x}\\
&\fragment{9}{= \left(\left( \frac{1}{\mathrm{sin}(z_1)} \right) \mathrm{cos}(x^2) \right) 2 x}\\
&\fragment{10}{= \left(\left( \frac{1}{\mathrm{sin}(x^2)} \right) \mathrm{cos}(x^2) \right) 2 x}\\
&\fragment{11}{= \left( \mathrm{cot}(x^2) \right) 2 x\ = 2 x ~\mathrm{tan} \left( \frac{\pi}{2} - x^2\right)}\\
\end{align}
</blockquote>
</section>
<section>
<h3>References</h3>
<ul style="list-style-type: none; font-size: 22pt">
<li class="fragment roll-in" data-fragment-index="0"> Appendix III of https://arxiv.org/pdf/1911.04048.pdf
<li class="fragment roll-in" data-fragment-index="0"> https://arxiv.org/pdf/1502.05767.pdf
<li class="fragment roll-in" data-fragment-index="0"> https://arxiv.org/pdf/1703.02311.pdf
</ul>
<div class="slide-footer">
<a href="verona_cnn.html">next section </a>
</div>
</section>
</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,
width: 960,
height: 700,
center: true,
hash: true,
controls: false,
keyboard: true,
margin: 0.05,
overview: true,
transition: 'slide', // Transition style: none/fade/slide/convex/concave/zoom
transitionSpeed: 'slow', // Transition speed: default/fast/slow
// hash: true,
// margin: 0.01,
// minScale: 0.01,
maxScale: 1.23,
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/mathjax@2.7.8/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"><h4>May 22, 2024</h4></div>
<div id="header-right"><h4></h4></div>
<div id="footer-left">
<img style="border:0; box-shadow: 0px 0px 0px rgba(150, 150, 255, 1);" width="200"
src="figures/UniverVerona.svg" alt="robot learning">
</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>