1
00:00:00,380 --> 00:00:01,980
Hello and welcome back.

2
00:00:01,980 --> 00:00:06,560
This week the first thing we'll do is show
you a number of case studies of the factor

3
00:00:06,560 --> 00:00:08,570
convolutional neural networks.

4
00:00:08,570 --> 00:00:10,330
So why look at case studies?

5
00:00:10,330 --> 00:00:13,480
Last week we learned about the basic
building blocks such as convolutional

6
00:00:13,480 --> 00:00:17,220
layers, proving layers and
fully connected layers of conv nets.

7
00:00:17,220 --> 00:00:21,320
It turns out a lot of the past few years
of computer vision research has been on

8
00:00:21,320 --> 00:00:23,910
how to put together these
basic building blocks

9
00:00:23,910 --> 00:00:26,770
to form effective
convolutional neural networks.

10
00:00:26,770 --> 00:00:27,990
And one of the best ways for

11
00:00:27,990 --> 00:00:32,220
you to get intuition yourself is
to see some of these examples.

12
00:00:32,220 --> 00:00:36,753
I think just as many of you may have
learned to write codes by reading other

13
00:00:36,753 --> 00:00:41,210
people's codes, I think that a good
way to get intuition on how to build

14
00:00:41,210 --> 00:00:46,140
conv nets is to read or to see other
examples of effective conv nets.

15
00:00:46,140 --> 00:00:50,270
And it turns out that a net neural
network architecture that works well on

16
00:00:50,270 --> 00:00:54,740
one computer vision task often works well
on other tasks as well such as maybe on

17
00:00:54,740 --> 00:00:55,750
your task.

18
00:00:55,750 --> 00:00:59,530
So if someone else is training neural
network as speak it out in your network

19
00:00:59,530 --> 00:01:03,012
architecture is very good at recognizing
cats and dogs and people but

20
00:01:03,012 --> 00:01:06,792
you have a different computer vision
task like maybe you're trying to sell

21
00:01:06,792 --> 00:01:07,900
self-driving car.

22
00:01:07,900 --> 00:01:11,867
You might well be able to take someone
else's neural network architecture and

23
00:01:11,867 --> 00:01:14,070
apply that to your problem.

24
00:01:14,070 --> 00:01:18,130
And finally, after the next few videos,
you'll be able to read some

25
00:01:18,130 --> 00:01:21,630
of the research papers from
the theater computer vision and

26
00:01:21,630 --> 00:01:24,515
I hope that you might find
it satisfying as well.

27
00:01:24,515 --> 00:01:28,545
You don't have to do this as a class but
I hope you might find it satisfying to be

28
00:01:28,545 --> 00:01:32,141
able to read some of these seminal
computer vision research paper and

29
00:01:32,141 --> 00:01:34,191
see yourself able to understand them.

30
00:01:34,191 --> 00:01:36,634
So with that, let's get started.

31
00:01:36,634 --> 00:01:40,711
As an outline of what we'll
do in the next few videos,

32
00:01:40,711 --> 00:01:44,256
we'll first show you
a few classic networks.

33
00:01:44,256 --> 00:01:48,663
The LeNEt-5 network which came from,
I guess, in 1980s,

34
00:01:48,663 --> 00:01:52,108
AlexNet which is often cited and
the VGG network and

35
00:01:52,108 --> 00:01:56,050
these are examples of pretty
effective neural networks.

36
00:01:56,050 --> 00:02:00,550
And some of the ideas lay the foundation
for modern computer vision.

37
00:02:00,550 --> 00:02:05,640
And you see ideas in these papers that
are probably useful for your own.

38
00:02:06,820 --> 00:02:10,340
And you see ideas from these papers
that were probably be useful for

39
00:02:10,340 --> 00:02:12,520
your own work as well.

40
00:02:12,520 --> 00:02:17,960
Then I want to show you the ResNet or
conv residual network and

41
00:02:17,960 --> 00:02:21,190
you might have heard that neural
networks are getting deeper and deeper.

42
00:02:21,190 --> 00:02:23,698
The ResNet neural network trained a very,

43
00:02:23,698 --> 00:02:28,439
very deep 152-layer neural network
that has some very interesting tricks,

44
00:02:28,439 --> 00:02:32,070
interesting ideas how
to do that effectively.

45
00:02:32,070 --> 00:02:38,720
And then finally you also see a case
study of the Inception neural network.

46
00:02:38,720 --> 00:02:43,436
After seeing these neural networks, l
think you have much better intuition about

47
00:02:43,436 --> 00:02:46,745
how to built effective
convolutional neural networks.

48
00:02:46,745 --> 00:02:49,947
And even if you end up not
working computer vision yourself,

49
00:02:49,947 --> 00:02:53,295
I think you find a lot of the ideas
from some of these examples,

50
00:02:53,295 --> 00:02:57,665
such as ResNet Inception network,
many of these ideas are cross-fertilizing

51
00:02:57,665 --> 00:03:00,105
on making their way
into other disciplines.

52
00:03:00,105 --> 00:03:03,715
So even if you don't end up building
computer vision applications yourself, I

53
00:03:03,715 --> 00:03:06,925
think you'll find some of these ideas very
interesting and helpful for your work.