1
00:00:00,000 --> 00:00:02,055
Congratulations on coming to

2
00:00:02,055 --> 00:00:03,930
the end of this first course,

3
00:00:03,930 --> 00:00:05,860
and well, you've come a long way.

4
00:00:05,860 --> 00:00:08,070
Yeah, I mean, we've
looked at with

5
00:00:08,070 --> 00:00:10,485
the students like you looked
at how to use TensorFlow,

6
00:00:10,485 --> 00:00:12,120
all the way from
the beginning with

7
00:00:12,120 --> 00:00:14,070
doing something like
fitting a straight line,

8
00:00:14,070 --> 00:00:16,755
to then like recognizing
fashion items.

9
00:00:16,755 --> 00:00:18,840
It's really cool. It's
like three lines of

10
00:00:18,840 --> 00:00:21,135
code to do a really
complicated task like that.

11
00:00:21,135 --> 00:00:22,260
Isn't it amazing, computer

12
00:00:22,260 --> 00:00:23,520
vision in three lines like that?

13
00:00:23,520 --> 00:00:26,235
Then, we improve that
by adding convolutions.

14
00:00:26,235 --> 00:00:27,510
We call it an improvement,

15
00:00:27,510 --> 00:00:29,160
but that means now
four lines of code to

16
00:00:29,160 --> 00:00:31,920
define two conv layers
and two maxpool layers,

17
00:00:31,920 --> 00:00:33,000
and then a two like

18
00:00:33,000 --> 00:00:35,190
TensorFlow take cares
of a lot of the rest.

19
00:00:35,190 --> 00:00:37,010
Then, we looked at going beyond

20
00:00:37,010 --> 00:00:40,210
the simple images to
bigger and colored images,

21
00:00:40,210 --> 00:00:42,365
and we didn't even add any lines
of code for that, right?

22
00:00:42,365 --> 00:00:43,610
It was just changing the shape of

23
00:00:43,610 --> 00:00:45,130
the data coming in. Yeah.

24
00:00:45,130 --> 00:00:47,420
So by learning
a framework like this,

25
00:00:47,420 --> 00:00:49,400
you can write
very complicated programs

26
00:00:49,400 --> 00:00:50,900
with just a few lines of code,

27
00:00:50,900 --> 00:00:53,345
and this helps people
save a lot of time.

28
00:00:53,345 --> 00:00:54,995
But there's still
a lot more to learn.

29
00:00:54,995 --> 00:00:56,900
Oh there is, like some
of the things that we'd

30
00:00:56,900 --> 00:00:58,850
love to explore are
things such as,

31
00:00:58,850 --> 00:01:00,680
when you have
very small data sets that

32
00:01:00,680 --> 00:01:02,630
can lead to an error
called over-fitting.

33
00:01:02,630 --> 00:01:03,860
So we'll explore
some techniques and

34
00:01:03,860 --> 00:01:06,310
tools that we can
use to avoid that.

35
00:01:06,310 --> 00:01:09,240
Transfer learning.
If you can download

36
00:01:09,240 --> 00:01:11,685
someone else's say
TensorFlow model,

37
00:01:11,685 --> 00:01:13,890
and use that for
your own problem,

38
00:01:13,890 --> 00:01:15,000
even though I was trained on

39
00:01:15,000 --> 00:01:16,620
a totally different data set,

40
00:01:16,620 --> 00:01:18,190
TensorFlow has tools that

41
00:01:18,190 --> 00:01:19,660
let you do that
efficiently as well.

42
00:01:19,660 --> 00:01:22,190
Absolutely, and one of
the things that a lot of

43
00:01:22,190 --> 00:01:23,630
AI practitioners love to be

44
00:01:23,630 --> 00:01:26,075
involved in is like
this online competitions,

45
00:01:26,075 --> 00:01:30,320
where places like Kango
provide a data set,

46
00:01:30,320 --> 00:01:31,640
and they ask you to
build a classifier

47
00:01:31,640 --> 00:01:32,730
around that data set.

48
00:01:32,730 --> 00:01:34,700
We're going to explore some
of the skills that you can

49
00:01:34,700 --> 00:01:37,105
use to take part in
those competitions.

50
00:01:37,105 --> 00:01:40,490
So congratulations again
on coming to the end of

51
00:01:40,490 --> 00:01:42,050
the first course
and there's still

52
00:01:42,050 --> 00:01:45,095
all these amazing things
to learn about TensorFlow.

53
00:01:45,095 --> 00:01:48,300
So please go on to
the next course.