1
00:00:00,000 --> 00:00:06,514
A few more important points about belief
networks, which are essentially Bayesian

2
00:00:06,514 --> 00:00:14,234
networks and their generalizations.
So far, we have talked about using,

3
00:00:14,234 --> 00:00:22,339
Bayesian networks to model simple
situations like, the sprinkler, rain and,

4
00:00:22,339 --> 00:00:27,780
wetness, etc., as well as to learn facts
from text.

5
00:00:28,460 --> 00:00:32,096
But, we haven't asked where these networks
come from.

6
00:00:32,096 --> 00:00:37,690
So far, we have used networks which we
have imagined from intuition or judgement.

7
00:00:37,690 --> 00:00:43,075
It turns out that, not only can the
conditional probabilities in the networks

8
00:00:43,075 --> 00:00:48,880
can be learned from data, but the network
structure itself can be learned from data.

9
00:00:48,880 --> 00:00:54,932
One of the greatest examples of learning
network structure from large volumes of

10
00:00:54,932 --> 00:00:59,415
data has been in genomic medicine, or in
medicine in general.

11
00:00:59,415 --> 00:01:04,870
Medical diagnosis, for example.
What treatments are best for what symptoms

12
00:01:04,870 --> 00:01:10,848
has been around for many years, and has
used Bayesian networks very successfully

13
00:01:10,848 --> 00:01:16,676
to build assisted systems for medical
diagnostics, especially in regions where

14
00:01:16,676 --> 00:01:23,458
there aren't that many qualified doctors.
In genomic medicine, the relationship

15
00:01:23,458 --> 00:01:29,561
between different genes expressing
themselves in an organism has been learned

16
00:01:29,561 --> 00:01:35,741
from large volumes of experimental data
using, using techniques for learning the

17
00:01:35,741 --> 00:01:41,303
structure of Bayesian networks.
Similarly, how phenotypes that is traits

18
00:01:41,303 --> 00:01:47,954
that are exhibited in an organism arise
from the genes of that organism, has also

19
00:01:47,954 --> 00:01:55,660
been, learned Bayesian networks which can
be inferred from large volumes of data.

20
00:01:56,820 --> 00:02:03,560
When it comes to logic and uncertainty,
there is a growing realization that belief

21
00:02:03,560 --> 00:02:09,889
networks are really bridging the gap
between the fundamental limits of logic

22
00:02:09,889 --> 00:02:16,281
and the pitfalls of uncertainty.
The indication of this is the fact that

23
00:02:16,281 --> 00:02:22,517
Judea Pearl, who invented Bayesian
networks was given the Turing award, which

24
00:02:22,517 --> 00:02:26,538
is the highest award in Computer Science,
in 2012.

25
00:02:26,538 --> 00:02:32,855
His initial work on Bayesian networks was
in the early 90s' and in fact, he's

26
00:02:32,855 --> 00:02:39,419
written a recent book on causality, which
is still a deep subject, not completely

27
00:02:39,419 --> 00:02:43,440
covered or even explored using Bayesian
networks.

28
00:02:43,760 --> 00:02:50,945
Other kinds of networks that merge logic
and probability are Markov logic networks,

29
00:02:50,945 --> 00:02:54,110
conditional learning fields, and many
others.

30
00:02:54,110 --> 00:03:00,954
We won't have time to even touch these in
this course, but they are all forms of

31
00:03:00,954 --> 00:03:06,600
belief networks that bridge the gap
between logic and uncertainty.

32
00:03:08,580 --> 00:03:13,281
Coming to big data,
Well, we have seen that inference in such

33
00:03:13,281 --> 00:03:20,726
networks can be done using SQL We've shown
this for Bayesian networks but the fact is

34
00:03:20,726 --> 00:03:24,801
that it's all counting and map-reduce
actually works.

35
00:03:24,801 --> 00:03:31,148
Because if we're just doing counting in
SQL, we can actually do inference from

36
00:03:31,148 --> 00:03:35,380
large volumes of data using the big data
technologies.

37
00:03:37,360 --> 00:03:45,370
As regards our hidden agenda about AI,
Deep belief networks, which we will study

38
00:03:45,370 --> 00:03:51,510
a little bit in the last week of lectures,
is one direction in which connectionist

39
00:03:51,510 --> 00:03:57,800
models of the brain are being explored and
extended, which sort of brings everything

40
00:03:57,800 --> 00:04:03,416
back together, that all the things that
we're studying in probability, statistics,

41
00:04:03,416 --> 00:04:09,481
Bayesian networks learning, eventually is
teaching us more and more about how the

42
00:04:09,481 --> 00:04:13,117
brain works.
This has been a long lecture.

43
00:04:13,117 --> 00:04:18,208
We've covered a lot of ground.
And it's worth recapping what we've

44
00:04:18,208 --> 00:04:23,387
actually learned.
First, we began by saying that search is

45
00:04:23,387 --> 00:04:29,988
not enough for general question and
answering on the web, which lead us to

46
00:04:29,988 --> 00:04:34,775
reasoning.
Logic and the semantic web vision is one

47
00:04:34,775 --> 00:04:41,284
way of trying to address this problem of
how computationally a web intelligence

48
00:04:41,284 --> 00:04:45,840
system could actually answer a general
purpose question.

49
00:04:46,500 --> 00:04:51,511
We learned, of course, that there are
fundamental limits to logic as well as

50
00:04:51,511 --> 00:04:56,990
practical ones arising from uncertainty.
We went into studying how reasoning under

51
00:04:56,990 --> 00:05:01,735
uncertainty could be handled using
Bayesian networks and probabilistic

52
00:05:01,735 --> 00:05:06,078
graphical models in general.
Though we didn't cover the latter, we

53
00:05:06,078 --> 00:05:09,620
indicated the direction in which this
field is going.

54
00:05:10,600 --> 00:05:16,660
In the next few weeks, we will have a
programming assignment next week.

55
00:05:17,560 --> 00:05:22,540
This will be on Bayesian inference using
SQL.

56
00:05:22,920 --> 00:05:31,096
We will have a short lecture video next
week to explain the assignment, but do

57
00:05:31,096 --> 00:05:40,158
start preparing by studying the SQL-based
inference that we've done in this week as

58
00:05:40,158 --> 00:05:46,670
well as experimenting with a sequel engine
of your choice, I would suggest SQLITE3,

59
00:05:46,670 --> 00:05:52,944
SQL Lite three, which comes bundled with
Python, and for which one can use an

60
00:05:52,944 --> 00:05:59,536
in-memory data base which will pretty much
suffice, since the tables we'll be using

61
00:05:59,536 --> 00:06:04,681
will be quite small.
The final week, will have the predict

62
00:06:04,681 --> 00:06:10,207
lecture, where we'll put everything
together, as well as have our final

63
00:06:10,207 --> 00:06:16,444
programming assignment and then the last
week, where there'll be the final exam,

64
00:06:16,444 --> 00:06:21,260
and you'll be asked to complete all your
assignments by then.

65
00:06:23,100 --> 00:06:30,836
So, do complete the homework and quiz for
this week and start preparing for Bayesian

66
00:06:30,836 --> 00:06:37,408
inference using SQL and experimenting with
SQL Lite for the next programming

67
00:06:37,408 --> 00:06:40,974
assignment.
The last week, which will have the final

68
00:06:40,974 --> 00:06:46,847
exam, and the submission dates for all the
programming assignments from here onwards.

69
00:06:46,847 --> 00:06:52,162
So, please prepare for next week's
programming assignment and do remember to

70
00:06:52,162 --> 00:06:57,476
complete the quiz and homework for this
week, which are due only next Friday,

71
00:06:57,476 --> 00:07:00,623
since we don't have a full lecture next
week.