1
00:00:00,000 --> 00:00:08,140
As usual, let's recap what we've done this
week and preview next week's lecture.

2
00:00:09,000 --> 00:00:16,794
We focused on learning or extracting
information from data such as figuring out

3
00:00:16,794 --> 00:00:24,266
classes from data in an unsupervised
manner using clustering. Figuring out

4
00:00:24,266 --> 00:00:28,700
rules from data using unsupervised rule
mining.

5
00:00:28,868 --> 00:00:36,475
And we also took a look at big data, and
how just counting techniques work well if

6
00:00:36,475 --> 00:00:43,511
you actually have lots of data, and our
unified f of x formulation helped up

7
00:00:43,511 --> 00:00:50,637
understand clustering, rule mining as well
as classification in the same single

8
00:00:50,637 --> 00:00:55,250
formulation.
We then turn to long data, high

9
00:00:55,250 --> 00:01:03,161
dimensional data, and figure out how learn
classes and features in an unsupervised

10
00:01:03,161 --> 00:01:12,444
model using hidden or latent techniques.
Next week, we'll continue our discussion

11
00:01:12,444 --> 00:01:19,255
of learning a little bit because the
techniques will be very similar to the

12
00:01:19,255 --> 00:01:23,466
others that we'll talk about next week as
well.

13
00:01:23,466 --> 00:01:30,546
So, we'll discuss learning facts from
collections of text via Bayesian networks

14
00:01:30,546 --> 00:01:38,167
and hidden Markov models returning once
more to supervised learning in a different

15
00:01:38,167 --> 00:01:44,997
form, not just classification.
And then, ask what use such rules and

16
00:01:44,997 --> 00:01:51,692
facts are if in fact one doesn't believe
John Sterling and one believes that one

17
00:01:51,692 --> 00:01:58,213
can reason using rules and facts to
connect the dots and make sense of the

18
00:01:58,213 --> 00:02:02,300
world just the way we put two and two
together.

19
00:02:03,740 --> 00:02:09,244
We'll talk about logical, as well as
probabilistic reasoning, reasoning under

20
00:02:09,244 --> 00:02:14,676
uncertainty, and most importantly, the
semantic web, where attempts have been

21
00:02:14,676 --> 00:02:20,615
made to put all these reasoning techniques
together in the context of data on the

22
00:02:20,615 --> 00:02:26,554
web, extracting facts and reasoning about
facts as being a higher order capability

23
00:02:26,554 --> 00:02:32,131
that today the web doesn't have. But,
might well do so in the very near future.

24
00:02:32,131 --> 00:02:33,652
So, see you next week.