1
00:00:00,000 --> 00:00:04,904
What is machine learning? In this video we
will try to define what it is and also try

2
00:00:04,904 --> 00:00:09,520
to give you a sense of when you want to
use machine learning. Even among machine

3
00:00:09,520 --> 00:00:14,252
learning practitioners there isn't a well
accepted definition of what is and what

4
00:00:14,252 --> 00:00:18,926
isn't machine learning. But let me show
you a couple of examples of the ways that

5
00:00:18,926 --> 00:00:23,600
people have tried to define it. Here's the
definition of what is machine learning

6
00:00:23,600 --> 00:00:28,520
does to Arthur Samuel. He defined machine
learning as the field of study that gives

7
00:00:31,037 --> 00:00:33,554
computers the ability to learn without being
explicitly programmed. Samuel's claim to

8
00:00:33,554 --> 00:00:38,452
fame was that back in the 1950's, he wrote
a checkers playing program. And the

9
00:00:38,452 --> 00:00:43,603
amazing thing about this checkers playing
program, was that Arthur Samuel himself,

10
00:00:43,603 --> 00:00:48,268
wasn't a very good checkers player. But
what he did was, he had to program for it to play

11
00:00:48,268 --> 00:00:52,245
10's of 1000's of games against itself.
And by watching what sorts of board

12
00:00:52,245 --> 00:00:56,698
positions tended to lead to wins, and what
sort of board positions tended to lead to

13
00:00:56,698 --> 00:01:00,725
losses. The checkers playing program
learns over time what are good board

14
00:01:00,725 --> 00:01:04,713
positions and what are bad board
positions. And eventually learn to play

15
00:01:04,713 --> 00:01:09,514
checkers better than Arthur Samuel himself
was able to. This was a remarkable result.

16
00:01:09,514 --> 00:01:14,535
Although Samuel himself turned out not to be a
very good checkers player. But because the

17
00:01:14,535 --> 00:01:19,254
computer has the patience to play tens
of thousands of games itself. No

18
00:01:19,254 --> 00:01:24,275
human, has the patience to play that many
games. By doing this the computer was able

19
00:01:24,275 --> 00:01:29,235
to get so much checkers-playing experience that it eventually became a

20
00:01:29,235 --> 00:01:33,817
better checkers player than Arthur Samuel
himself. This is somewhat informal

21
00:01:33,817 --> 00:01:38,547
definition, and an older one. Here's a
slightly more recent definition by Tom

22
00:01:38,547 --> 00:01:43,607
Mitchell, who's a friend out of Carnegie
Mellon. So Tom defines machine learning by

23
00:01:43,607 --> 00:01:48,819
saying that, a well posed learning problem
is defined as follows. He says, a computer

24
00:01:48,819 --> 00:01:53,843
program is said to learn from experience
E, with respect to some task T, and some

25
00:01:53,843 --> 00:01:58,678
performance measure P, if its
performance on T as measured by P improves

26
00:01:58,678 --> 00:02:03,764
with experience E. I actually think he came
up with this definition just to make it

27
00:02:03,764 --> 00:02:08,346
rhyme. For the checkers playing
example the experience e, will be the

28
00:02:08,346 --> 00:02:13,253
experience of having the program play 10's
of 1000's of games against itself. The

29
00:02:13,253 --> 00:02:17,735
task t, will be the task of playing
checkers. And the performance measure p,

30
00:02:17,735 --> 00:02:22,399
will be the probability that it
wins the next game of checkers against

31
00:02:22,399 --> 00:02:27,157
some new opponent. Throughout these
videos, besides me trying to teach you

32
00:02:27,157 --> 00:02:32,291
stuff, I will occasionally ask you a
question to make sure you understand the

33
00:02:32,291 --> 00:02:36,891
content. Here's one, on top is a
definition of machine learning by Tom

34
00:02:36,891 --> 00:02:42,292
Mitchell. Let's say your email program
watches which emails you do or do not flag

35
00:02:42,292 --> 00:02:47,826
as spam. So in an email client like this
you might click this spam button to report

36
00:02:47,826 --> 00:02:53,263
some email as spam, but not other emails
and. Based on which emails you mark as

37
00:02:53,263 --> 00:02:59,046
spam, so your e-mail program learns better
how to filter spam e-mail. What is the

38
00:02:59,046 --> 00:03:04,290
task T in this setting? In a few seconds,
the video will pause. And when it does so,

39
00:03:04,290 --> 00:03:09,598
you can use your mouse to select one of
these four radio buttons to let, to let me

40
00:03:09,598 --> 00:03:40,190
know which of these four you think is the
right answer to this question. That might

41
00:03:40,190 --> 00:03:45,747
be a performance measure P. And so, our
task performance on the task our system's

42
00:03:45,747 --> 00:03:50,529
performance on the task T, on the
performance measure P will improve after

43
00:03:50,529 --> 00:03:55,957
the experience E. In this class I hope to
teach you about various different types of

44
00:03:55,957 --> 00:04:00,933
learning algorithms. There are several
different types of learning algorithms.

45
00:04:00,933 --> 00:04:05,650
The main two types are what we call
supervised learning and unsupervised

46
00:04:05,650 --> 00:04:10,690
learning. I'll define what these terms
mean more in the next couple videos. But

47
00:04:10,690 --> 00:04:16,028
it turns out that in supervised learning,
the idea is that we're going to teach the

48
00:04:16,028 --> 00:04:20,513
computer how to do something, whereas in
unsupervised learning we're going let

49
00:04:20,513 --> 00:04:25,016
it learn by itself. Don't worry if these
two terms don't make sense yet, in the

50
00:04:25,016 --> 00:04:29,739
next two videos I'm going to say exactly
what these two types of learning are. You

51
00:04:29,739 --> 00:04:34,070
will also hear other buzz terms such as
reinforcement learning and recommender

52
00:04:34,070 --> 00:04:38,621
systems. These are other types of machine
learning algorithms that we'll talk about

53
00:04:38,621 --> 00:04:42,460
later but the two most used types of
learning algorithms are probably

54
00:04:42,460 --> 00:04:46,791
supervised learning and unsupervised
learning and I'll define them in the next

55
00:04:46,791 --> 00:04:51,123
two videos and we'll spend most of this
class talking about these two types of

56
00:04:51,123 --> 00:04:55,720
learning algorithms. It turns out one of
the other things we'll spend a lot of time

57
00:04:55,720 --> 00:05:00,054
on in this class is practical advice for
applying learning algorithms. This is

58
00:05:00,054 --> 00:05:04,444
something that I feel pretty strongly
about, and it's actually something that I

59
00:05:04,444 --> 00:05:08,167
don't know of any other university
teaches. Teaching about learning

60
00:05:08,167 --> 00:05:12,509
algorithms is like giving you a set of
tools, and equally important or more

61
00:05:12,509 --> 00:05:17,616
important to giving you the tools is to
teach you how to apply these tools. I like

62
00:05:17,616 --> 00:05:22,413
to make an analogy to learning to become a
carpenter. Imagine that someone is

63
00:05:22,413 --> 00:05:26,959
teaching you how to be a carpenter and
they say here's a hammer, here's a

64
00:05:26,959 --> 00:05:31,077
screwdriver, here's a saw, good luck.
Well, that's no good, right? You, you, you

65
00:05:31,077 --> 00:05:34,799
have all these tools, but the more
important thing, is to learn how to use

66
00:05:34,799 --> 00:05:38,927
these tools properly. There's a huge
difference between, between people that

67
00:05:38,927 --> 00:05:43,456
know how to use these machines learning
algorithms, versus people who don't know

68
00:05:43,456 --> 00:05:47,626
how to use these tools well. Here in
Silicon Valley where I live, when I go

69
00:05:47,626 --> 00:05:52,328
visit different companies even at the
top Silicon Valley companies very often I see

70
00:05:52,328 --> 00:05:56,428
people are trying to apply machine
learning algorithms to some problem and

71
00:05:56,428 --> 00:06:00,857
sometimes they have been going at it for
six months. But sometimes when I look at

72
00:06:00,857 --> 00:06:05,121
what they're doing I, I, I say, you know,
I could have told them like, gee, I could

73
00:06:05,121 --> 00:06:09,714
have told you six months ago that you
should be taking a learning algorithm and

74
00:06:09,714 --> 00:06:14,470
applying it in like the slightly modified
way and your chance of success would have

75
00:06:14,470 --> 00:06:19,648
been much higher. So what we're going to
do in this class is actually spend a lot

76
00:06:19,648 --> 00:06:23,523
of time talking about how, if you actually
tried to develop a machine learning

77
00:06:23,523 --> 00:06:27,596
system, how to make those best practices
type decisions about the way in which you

78
00:06:27,596 --> 00:06:31,321
build your system so that when you're
applying learning algorithm you're less

79
00:06:31,321 --> 00:06:35,394
likely to end up one of those people who
end up pursuing some path for six months

80
00:06:35,394 --> 00:06:39,373
that, you know, someone else could have
figured out it just wasn't gonna work at

81
00:06:39,373 --> 00:06:43,515
all and it's just a waste of time for six
months. So I'm actually going to spend a

82
00:06:43,515 --> 00:06:47,707
lot of the time teaching you those sorts
of best practices in machine learning and

83
00:06:47,707 --> 00:06:52,052
AI and how to get this stuff to work and
how we do it, how the best people do it in

84
00:06:52,052 --> 00:06:56,143
Silicon Valley and around the world. I
hope to make you one of the best people in

85
00:06:56,143 --> 00:06:59,905
knowing how to design and build serious
machine learning and AI systems. So,

86
00:06:59,905 --> 00:07:04,698
that's machine learning and these are the
main topics I hope to teach. In the next

87
00:07:04,698 --> 00:07:09,023
video, I'm going to define what is
supervised learning and after that, what

88
00:07:09,023 --> 00:07:13,757
is unsupervised learning. And also, start
to talk about when you would use each of them.