1
00:00:00,000 --> 00:00:06,019
Let's now look at some applications where
the relationships between prediction,

2
00:00:06,019 --> 00:00:12,344
making decisions based on the predictions,
and then controlling when actions become

3
00:00:12,344 --> 00:00:16,832
abundantly clear.
First example is Robo-soccer.

4
00:00:16,832 --> 00:00:22,477
I don't know if many of you have tried
this but, you know, the regular

5
00:00:22,477 --> 00:00:28,776
competitions of people building small
robots and getting them to play soccer.

6
00:00:28,776 --> 00:00:33,358
And this is a little photograph of one
such competition.

7
00:00:33,358 --> 00:00:40,070
Imagine what this robot has to do.
It needs to predict where the ball will

8
00:00:40,070 --> 00:00:43,274
be.
Decide the best path to get to the ball

9
00:00:43,274 --> 00:00:47,261
and navigate there.
And that's, that's the basic thing that it

10
00:00:47,261 --> 00:00:50,660
needs to do if it doesn't have position of
the ball.

11
00:00:51,960 --> 00:00:58,245
Prediction is very critical, the ball may
be moving, with a certain velocity, and

12
00:00:58,245 --> 00:01:03,098
it's and the robot need to figure out
where it's going to go.

13
00:01:03,098 --> 00:01:09,623
Much like our very first example of the
paddle game in the zeroth lecture of this

14
00:01:09,623 --> 00:01:14,250
course.
Another example is self-driving cars.

15
00:01:14,250 --> 00:01:20,638
You might have heard that Google, for
example, is developing a self-driving car,

16
00:01:20,638 --> 00:01:27,108
which has already driven over 100,000
kilometers both in rough terrain as well

17
00:01:27,108 --> 00:01:31,940
as in urban environments.
Imagine what such a car has to do.

18
00:01:31,940 --> 00:01:37,419
Needs to predict the path of a pedestrian
that it see.

19
00:01:37,419 --> 00:01:41,802
Decide the path it needs to follow to
avoid that person.

20
00:01:41,802 --> 00:01:46,889
And then steer the car appropriately.
Prediction, decision, that is

21
00:01:46,889 --> 00:01:50,960
optimization, best path.
And then control or steering.

22
00:01:52,660 --> 00:01:58,992
We've all heard about the intelligent
energy grid, which is going to change the

23
00:01:58,992 --> 00:02:02,920
way we consume electricity and natural
resources.

24
00:02:04,900 --> 00:02:11,010
These things happened already today.
Great predict of the energy demand, decide

25
00:02:11,010 --> 00:02:17,276
and control the distribution, all using
techniques very similar to those that we

26
00:02:17,276 --> 00:02:20,880
have discussed, and these are already in
place.

27
00:02:23,200 --> 00:02:28,468
Think about a business of having to deal
with a supply chain of products being

28
00:02:28,668 --> 00:02:33,869
built, produced based on raw materials
which have to be sourced from rather

29
00:02:33,869 --> 00:02:38,804
different locations around the world.
In order to decide which products to

30
00:02:38,804 --> 00:02:43,806
produce and what quantities, a supply
chain needs to predict the demand for

31
00:02:43,806 --> 00:02:47,440
every product.
Based on that, decide the best production

32
00:02:47,440 --> 00:02:52,645
plan based on the raw materials available
and which products require which raw

33
00:02:52,645 --> 00:02:57,322
materials in which quantities,
An optimization problem and then execute

34
00:02:57,322 --> 00:02:59,957
it.
A very complex control system indeed,

35
00:02:59,957 --> 00:03:04,963
which, which involves many, many moving
parts and many pieces of a complex

36
00:03:04,963 --> 00:03:10,970
organization.
Another example of supply chain.

37
00:03:10,970 --> 00:03:15,366
Is.
When products are manufactured all around

38
00:03:15,366 --> 00:03:18,473
the world, there could be a variety of
risks.

39
00:03:18,473 --> 00:03:24,759
You could have floods or fires or unstable
political conditions in different parts of

40
00:03:24,759 --> 00:03:28,926
the world which might risk, pose a risk to
one supply chain.

41
00:03:28,926 --> 00:03:33,800
These kinds of events can be detected
increasingly from social media.

42
00:03:33,800 --> 00:03:39,396
Sources such as Twitter, once such risks
are detected or potential risks are

43
00:03:39,396 --> 00:03:45,655
detected they need to be evaluated whether
or not they pose significant risks and if

44
00:03:45,655 --> 00:03:50,810
so, the production plan needs to be
re-planned and then re-executed or.

45
00:03:50,810 --> 00:03:54,220
Amendments to that plan need to be put
into place.

46
00:03:55,180 --> 00:04:00,119
Finally, in marketing.
When one is trying to sell products, one

47
00:04:00,119 --> 00:04:06,678
might need to predict the demand for a
product, decide the promotion strategy per

48
00:04:06,678 --> 00:04:12,103
region based on the demand and then
execute the promotion strategy.

49
00:04:12,103 --> 00:04:16,800
In each case we have the predict, decide
and control piece.

50
00:04:18,140 --> 00:04:25,295
These examples are not necessarily simple,
but they can get increasingly complicated

51
00:04:25,295 --> 00:04:32,480
as we shall see in a minute.
Think about the examples below, which are

52
00:04:32,480 --> 00:04:37,662
grayed out now.
And compare these to robo software self

53
00:04:37,662 --> 00:04:42,655
driving car energy grid.
They'd, there does appear to be something

54
00:04:42,655 --> 00:04:46,884
fundamentally different about the ones at
the bottom and the ones at top.

55
00:04:46,884 --> 00:04:51,525
For example, the ones on the top are
possibly executable or implementable

56
00:04:51,525 --> 00:04:54,978
using, say, a system which will predict
the ball.

57
00:04:54,978 --> 00:04:59,260
Directly take that prediction.
Decide the best path using some

58
00:04:59,260 --> 00:05:03,612
optimization techniques.
And then navigate there using a simple

59
00:05:03,612 --> 00:05:06,858
control system.
Similarly for self-driving cars.

60
00:05:07,065 --> 00:05:13,419
In fact, auto-pilots for airplanes are
already part and parcel of every day life.

61
00:05:13,626 --> 00:05:19,013
Cars are a little bit more complicated,
because of things like pedestrians and

62
00:05:19,013 --> 00:05:22,260
overtaking and lanes.
But still, one can imagine.

63
00:05:23,080 --> 00:05:29,089
Which is, you know, a suitably implemented
system actually doing this, and simulate

64
00:05:29,089 --> 00:05:34,505
the energy great as that said is already
there, and running already today.

65
00:05:34,505 --> 00:05:37,918
However, consider some of these examples
below.

66
00:05:37,918 --> 00:05:43,705
Detecting potential risks from say, tweets
is possible, but then evaluating the

67
00:05:43,705 --> 00:05:48,380
impact requires certain degree of fairly
complicated reasoning.

68
00:05:49,040 --> 00:05:54,976
Similarly, prediction demand might be easy
but deciding the right promotion strategy

69
00:05:54,976 --> 00:05:58,440
by region might be a fairly complicated
exercise.

70
00:05:58,740 --> 00:06:02,166
The problems above can also be made more
complicated.

71
00:06:02,166 --> 00:06:07,000
For example in addition to predicting
where the ball will be, one might want to

72
00:06:07,000 --> 00:06:11,528
predict how other players will move.
And this requires a certain degree of

73
00:06:11,528 --> 00:06:15,177
reasoning.
Similarly, one might need to predict

74
00:06:15,177 --> 00:06:21,217
traffic based on possible inputs from many
different people about where they want to

75
00:06:21,217 --> 00:06:24,992
go that day.
Based on that, you might want to decide

76
00:06:24,992 --> 00:06:28,814
the optimal roots for to destination for
everybody.

77
00:06:28,814 --> 00:06:34,585
And in some sense, a self-driving car
might manifest itself into a network of

78
00:06:34,585 --> 00:06:39,906
many, many self driving cars.
Along with people deciding where they want

79
00:06:39,906 --> 00:06:44,927
to go through their cell phones.
And all the cars magically getting

80
00:06:44,927 --> 00:06:48,900
everybody where they want to be with
minimal traffic.

81
00:06:49,160 --> 00:06:53,318
A fairly utopian situation but, in
principle, possible.

82
00:06:53,318 --> 00:06:58,340
Again, a little more complicated
decision-making needs to happen.

83
00:06:59,040 --> 00:07:03,560
Reasoning to a certain extend, a lot of
planning for sure.

84
00:07:03,980 --> 00:07:09,324
Similarly in the energy grid, you might
want to predict the supply by how green

85
00:07:09,324 --> 00:07:13,247
that supply is.
Whether it's coming from a wind turbine or

86
00:07:13,247 --> 00:07:18,659
from a coal fired power plant, and adjust
the prices optimally so that one could,

87
00:07:18,862 --> 00:07:24,138
maximize the utility of the green power
when it's available, that is, let's say

88
00:07:24,138 --> 00:07:29,685
when the wind is blowing, but, not,
necessarily, make people pay too much if

89
00:07:29,685 --> 00:07:34,150
there's no alternative to a coal fired
power at a particular time.

90
00:07:34,150 --> 00:07:38,724
Again, more complicated predict the site
control systems.

91
00:07:38,724 --> 00:07:45,341
All these are, today being built or are
already in production to a certain extent.

92
00:07:45,341 --> 00:07:52,040
And because of the number of devices, the
number of censors and information on the

93
00:07:52,040 --> 00:07:56,451
web, on social media.
And the interconnection of almost

94
00:07:56,451 --> 00:08:01,597
everything in general.
I definitely call all of these potential

95
00:08:01,597 --> 00:08:06,076
web intelligence systems.
They incorporate many of the techniques

96
00:08:06,076 --> 00:08:11,132
that we have talked about, as well as many
that we haven't such as, optimization and

97
00:08:11,132 --> 00:08:15,937
control.
Well, having seen a number of applications

98
00:08:16,167 --> 00:08:22,143
this lecture we are talking about
prediction and, and certainly learning, so

99
00:08:22,143 --> 00:08:28,272
we will now briefly come back to the
techniques and try to see which techniques

100
00:08:28,272 --> 00:08:30,954
one should use in which situations.