1
00:00:00,000 --> 00:00:03,579
[MUSIC]

2
00:00:03,579 --> 00:00:06,690
So how did you get into machine learning?

3
00:00:06,690 --> 00:00:07,200
>> It's interesting.

4
00:00:07,200 --> 00:00:09,310
So as I talked about
earlier in the module,

5
00:00:09,310 --> 00:00:12,480
I started because I really
cared about robots.

6
00:00:12,480 --> 00:00:14,400
I really wanted to build robots,
and I thought okay,

7
00:00:14,400 --> 00:00:17,100
the way to build robots is to
do mechanical engineering.

8
00:00:17,100 --> 00:00:20,340
So I started mechanical engineering, and
then I realized the hard thing about

9
00:00:20,340 --> 00:00:24,080
building robots is the robot brains, and
that took me to computer science, and

10
00:00:24,080 --> 00:00:27,428
then I realized the hard part of
the AI was understanding data and

11
00:00:27,428 --> 00:00:31,088
they got me into machine learning so
I did my PhD in machine learning.

12
00:00:31,088 --> 00:00:33,770
So that's the journey from
thinking about robots,

13
00:00:33,770 --> 00:00:38,130
all the way to becoming somebody excited
about data and about machine learning.

14
00:00:38,130 --> 00:00:40,650
So how about you?

15
00:00:40,650 --> 00:00:41,770
>> Similar but different journey.

16
00:00:41,770 --> 00:00:47,740
I wasn't sitting there reading Issac
Asimov making these lifetime aspirations

17
00:00:47,740 --> 00:00:51,080
to build robots, but eventually I
actually was working with robots.

18
00:00:52,080 --> 00:00:55,310
But the thing that really got me
into machine learning is I was

19
00:00:55,310 --> 00:00:59,070
looking on this target tracking
application where they're

20
00:00:59,070 --> 00:01:04,290
trying to track these really highly
evasive vehicles like a plane.

21
00:01:04,290 --> 00:01:09,060
So a plane is flying and make some
maneuver to try and avoid being tracked.

22
00:01:09,060 --> 00:01:12,440
But the issue is that
the systems were really,

23
00:01:12,440 --> 00:01:17,530
really highly engineered to assuming that
you knew the physics of this airplane,

24
00:01:17,530 --> 00:01:21,290
that you knew that it was gonna do some
kind of turn right maneuver and turn left.

25
00:01:21,290 --> 00:01:26,070
And this is how the physics are when
the plane is flying straight or

26
00:01:26,070 --> 00:01:27,200
landing or things like this.

27
00:01:27,200 --> 00:01:33,130
And so you would build these models that
were known based on physics of the system.

28
00:01:33,130 --> 00:01:34,440
All these different modes but

29
00:01:34,440 --> 00:01:38,600
what would happen is the thing would do
something that we hadn't seen before and

30
00:01:38,600 --> 00:01:42,040
the system would try to accommodate for
it by saying, oh well,

31
00:01:42,040 --> 00:01:45,320
it's just rapidly switching between turn
right, turn left, turn right, turn left.

32
00:01:45,320 --> 00:01:47,880
But, of course, that didn't actually
describe what was going on, so

33
00:01:47,880 --> 00:01:52,560
I thought there must be some better
way to actually learn the parameters

34
00:01:52,560 --> 00:01:57,190
of the models that we are trying
to deploy And somehow capture

35
00:01:57,190 --> 00:02:02,440
uncertainty about how much we actually
believe in the model that we specified,

36
00:02:02,440 --> 00:02:05,650
to be a bit more robust to
these types of situations.

37
00:02:05,650 --> 00:02:08,670
So, that's what got me into
machine learning and statistics.

38
00:02:09,750 --> 00:02:13,270
>> Right, so as you can see,
I came from mechanical engineering and

39
00:02:13,270 --> 00:02:15,200
got into data and to machine learning.

40
00:02:15,200 --> 00:02:17,930
Came from electrical engineering and
tracking planes and

41
00:02:17,930 --> 00:02:19,580
got into machine learning.

42
00:02:19,580 --> 00:02:23,370
But today, it's really exciting because
we see people from all sorts of fields

43
00:02:23,370 --> 00:02:24,190
get in to machine learning.

44
00:02:24,190 --> 00:02:27,210
We see physicists excited about data.

45
00:02:27,210 --> 00:02:29,630
We see people in biology.

46
00:02:29,630 --> 00:02:32,700
>> Social sciences, philosophy.

47
00:02:32,700 --> 00:02:36,839
>> All sorts of place, math,
we do self statistics,

48
00:02:36,839 --> 00:02:39,350
we do science, all the
>> Art.

49
00:02:39,350 --> 00:02:40,670
You always like to mention art.

50
00:02:40,670 --> 00:02:42,410
>> Oh, yeah.
When I talked at Carnegie Mellon

51
00:02:42,410 --> 00:02:45,220
there were some art students that
took my machine learning class.

52
00:02:45,220 --> 00:02:49,020
And did really cool art installations
that use machine learning inside them,

53
00:02:49,020 --> 00:02:50,410
which is really quite amazing.

54
00:02:50,410 --> 00:02:54,040
We even taught a class
there with my friend

55
00:02:54,040 --> 00:02:56,620
on building art installations
that use machine language.

56
00:02:56,620 --> 00:03:00,960
So we see a variety of people with
different backgrounds, staging the lives,

57
00:03:00,960 --> 00:03:03,466
thinking about data science,
thinking about machine learning and

58
00:03:03,466 --> 00:03:05,930
building intelligent applications and
they all have a passion for data.

59
00:03:05,930 --> 00:03:07,640
Which is cool.

60
00:03:08,740 --> 00:03:11,979
>> It is cool, that's how we're
gonna end every one of these chats.

61
00:03:11,979 --> 00:03:12,735
That's cool.

62
00:03:12,735 --> 00:03:13,536
It is cool.

63
00:03:13,536 --> 00:03:14,371
>> It is cool.

64
00:03:14,371 --> 00:03:17,693
[LAUGH]
>> See you next chat.

65
00:03:17,693 --> 00:03:19,539
>> Cool [LAUGH]

66
00:03:19,539 --> 00:03:23,754
[MUSIC]