1
00:00:00,000 --> 00:00:03,210
So in that lesson, we
just saw the basics of

2
00:00:03,210 --> 00:00:04,830
the new programming paradigm that

3
00:00:04,830 --> 00:00:06,630
comes with machine learning
and deep learning,

4
00:00:06,630 --> 00:00:08,010
and how instead of like

5
00:00:08,010 --> 00:00:10,200
expressing rules in
a programming language,

6
00:00:10,200 --> 00:00:11,970
we can start getting
data and using

7
00:00:11,970 --> 00:00:13,530
labeled data to open up

8
00:00:13,530 --> 00:00:15,630
new scenarios like
activity recognition.

9
00:00:15,630 --> 00:00:17,040
Then for a little bit of fun,

10
00:00:17,040 --> 00:00:19,080
we actually started doing
our first piece of code.

11
00:00:19,080 --> 00:00:22,620
We built a super simple neural
network that fit data like

12
00:00:22,620 --> 00:00:24,120
an x and y data onto a line

13
00:00:24,120 --> 00:00:26,595
but that was just "Hello,
World". Right, Andrew?

14
00:00:26,595 --> 00:00:29,370
So fitting straight lines
seems like the "Hello,

15
00:00:29,370 --> 00:00:32,130
world" most basic implementation
learning algorithm.

16
00:00:32,130 --> 00:00:33,930
But one of the most
amazing things

17
00:00:33,930 --> 00:00:35,370
about machine learning is that,

18
00:00:35,370 --> 00:00:37,890
that core of the idea
of fitting the x and

19
00:00:37,890 --> 00:00:41,445
y relationship is what lets
us do amazing things like,

20
00:00:41,445 --> 00:00:42,960
have computers look
at the picture

21
00:00:42,960 --> 00:00:44,565
and do activity recognition,

22
00:00:44,565 --> 00:00:46,560
or look at the picture
and tell us,

23
00:00:46,560 --> 00:00:48,330
is this a dress,

24
00:00:48,330 --> 00:00:49,880
or a pair of pants,
or a pair of shoes

25
00:00:49,880 --> 00:00:51,500
really hard for humans,

26
00:00:51,500 --> 00:00:53,210
and amazing that computers can

27
00:00:53,210 --> 00:00:55,250
now use this to do
these things as well.

28
00:00:55,250 --> 00:00:56,360
Right, like computer vision is

29
00:00:56,360 --> 00:00:58,070
a really hard problem
to solve, right?

30
00:00:58,070 --> 00:00:59,840
Because you're saying
like dress or shoes.

31
00:00:59,840 --> 00:01:01,460
It's like how would I
write rules for that?

32
00:01:01,460 --> 00:01:04,100
How would I say, if this
pixel then it's a shoe,

33
00:01:04,100 --> 00:01:05,300
if that pixel then its a dress.

34
00:01:05,300 --> 00:01:06,655
It's really hard to do, so

35
00:01:06,655 --> 00:01:08,825
the labeled samples are
the right way to go.

36
00:01:08,825 --> 00:01:12,320
Yeah. One of the non-intuitive
things about vision is

37
00:01:12,320 --> 00:01:15,905
that it's so easy for a person
to look at you and say,

38
00:01:15,905 --> 00:01:17,630
you're wearing a shirt,
it's so hard for

39
00:01:17,630 --> 00:01:19,930
a computer to figure it out.

40
00:01:19,930 --> 00:01:23,190
Because it's so easy for
humans to recognize objects,

41
00:01:23,190 --> 00:01:25,280
it's almost difficult
to understand why this

42
00:01:25,280 --> 00:01:27,815
is a complicated thing
for a computer to do.

43
00:01:27,815 --> 00:01:30,815
What the computer has to
do is look at all numbers,

44
00:01:30,815 --> 00:01:32,240
all the pixel brightness value,

45
00:01:32,240 --> 00:01:33,560
saying look at all of
these numbers saying,

46
00:01:33,560 --> 00:01:36,170
these numbers correspond
to a black shirt,

47
00:01:36,170 --> 00:01:38,230
and it's amazing that
with machine and

48
00:01:38,230 --> 00:01:41,120
deep learning computers are
getting really good at this.

49
00:01:41,120 --> 00:01:43,235
Right, so it's like with

50
00:01:43,235 --> 00:01:44,780
the code that we just used in

51
00:01:44,780 --> 00:01:46,610
the previous lesson
as you mentioned,

52
00:01:46,610 --> 00:01:49,640
it provides a template for
everything that we can do with

53
00:01:49,640 --> 00:01:51,890
deep learning by designing
a Neural network in

54
00:01:51,890 --> 00:01:54,170
the layers to be able to
recognize patterns like this.

55
00:01:54,170 --> 00:01:56,000
So maybe we can do that with

56
00:01:56,000 --> 00:01:58,620
clothes recognition
today. What do you think?

57
00:01:58,620 --> 00:02:00,510
Yeah. So in the next video,

58
00:02:00,510 --> 00:02:03,330
you'll learn how to
write code to take

59
00:02:03,330 --> 00:02:06,735
this paradigm you've already
saw in a previous video,

60
00:02:06,735 --> 00:02:08,135
but to now apply it to

61
00:02:08,135 --> 00:02:10,505
recognizing clothes
from labeled data.

62
00:02:10,505 --> 00:02:12,810
Please go on to the next video.