-
Notifications
You must be signed in to change notification settings - Fork 25
/
Copy path03-00-hands-on.html
141 lines (96 loc) · 2.95 KB
/
03-00-hands-on.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
<!DOCTYPE html>
<html>
<head>
<title>Hands On</title>
<meta charset="utf-8">
<style>
@import url(https://fonts.googleapis.com/css?family=Montserrat);
@import url(https://fonts.googleapis.com/css?family=Lato:400,700,400italic);
@import url(https://fonts.googleapis.com/css?family=Source+Code+Pro:400,700,400italic);
body { font-family: 'Lato'; }
h1, h2, h3 {
font-family: 'Montserrat';
font-weight: normal;
}
img {
max-width: 100%;
}
video {
max-width: 100%;
}
.remark-code, .remark-inline-code { font-family: 'Source Code Pro'; }
</style>
</head>
<body>
<textarea id="source">
class: center, middle
# Hands On
### Adversarial examples
### Transfer learning
???
* One of the challenges here is that most cool stuff requires tons of compute
* If you're serious about this stuff, buy yourself a nice GPU, and try to get
your hands on a ton of cloud compute credit (or join a lab or something)
* We have two areas for you to explore that actually don't need that much
compute
---
# Adversarial examples
* Neural networks are easily fooled!
--
.center[
]
???
* Both a theoretical and practical concern. Theoretically: why is it that these
networks seem to generalize well, but fail on these adversarial examples that
are imperceptibly perturbed images? Practically: are real systems at risk?
* [ 03-01-notes ]
* Bunch of neat results in this area
---
# Adversarial examples
* Can even have adversarial objects
--
.center[
<video src="figs/turtle.mp4" autoplay="" loop="" controls="" style="width: 100%;">
</video>
]
???
* This is what the adversarial examples notebook is all about. What's neat is
that it's not that computationally hard to do this, at least for two
dimensional images: you can even do it with your CPU on your laptop.
---
# Transfer learning
* Easy way to train networks to solve custom tasks without requiring tons of compute
???
* [ 03-02-notes ]
--
.center[
]
* (play it yourself: https://tenso.rs/demos/rock-paper-scissors/)
???
* This is what the transfer learning notebook is all about.
---
???
* Blank slide
* Briefly go through notebooks / code
* See interest split, split people between rooms
* Work in small groups
* Ask people to come back for wrap-up
</textarea>
<script src="https://gnab.github.io/remark/downloads/remark-latest.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.1/MathJax.js?config=TeX-AMS_HTML&delayStartupUntil=configured" type="text/javascript"></script>
<script type="text/javascript">
var slideshow = remark.create({
countIncrementalSlides: false
});
// Setup MathJax
MathJax.Hub.Config({
tex2jax: {
skipTags: ['script', 'noscript', 'style', 'textarea', 'pre']
}
});
MathJax.Hub.Configured();
</script>
</body>
</html>