1
00:00:00,000 --> 00:00:04,572
[MUSIC]

2
00:00:04,572 --> 00:00:08,540
Let's take a final minute to summarize
what we have seen in today's module.

3
00:00:08,540 --> 00:00:14,626
Around scaling to large datasets and
online learning.

4
00:00:14,626 --> 00:00:17,873
Now, we use gradient to
scale to large datasets, but

5
00:00:17,873 --> 00:00:20,570
this is not the only approach that's used.

6
00:00:20,570 --> 00:00:26,590
In fact, a lot of scaling that we do is
about dealing with multicore processors.

7
00:00:26,590 --> 00:00:30,840
So, the fact that processors have multiple
cores and it can do parallel computation.

8
00:00:30,840 --> 00:00:35,361
And then, using large computers clusters
with 10 of machines, 100 machines,

9
00:00:35,361 --> 00:00:39,485
1,000 machines,
100,000 machines, and so on.

10
00:00:39,485 --> 00:00:43,435
And this requires new kinds of machinery
algorithms which are called distributed

11
00:00:43,435 --> 00:00:47,855
and parallel algorithms that can be
spread out over a thousand machine.

12
00:00:47,855 --> 00:00:49,115
It's a very important topic.

13
00:00:49,115 --> 00:00:52,250
It's actually what a lot of my
personal research is about.

14
00:00:52,250 --> 00:00:56,595
How do you do this written machine or any
algorithms, unfortunately is not something

15
00:00:56,595 --> 00:00:59,630
that are not going to have time
to cover in the current course.

16
00:01:00,810 --> 00:01:04,970
As a summary we seen that a small
modification of your gradient

17
00:01:04,970 --> 00:01:09,976
algorithm can incredible improvement in
the overall running time of the approach,

18
00:01:09,976 --> 00:01:13,100
and that makes a huge
difference in practice.

19
00:01:13,100 --> 00:01:17,910
The many practical challenges associated
in gradient is extremely useful.

20
00:01:17,910 --> 00:01:21,950
And the same kind of techniques that
we saw here can also be used for

21
00:01:21,950 --> 00:01:26,504
online learning which is a whole
different type of machine learning which

22
00:01:26,504 --> 00:01:29,159
is also of great practical significance.

23
00:01:29,159 --> 00:01:33,069
[MUSIC]