1
00:00:11,340 --> 00:00:16,020
So in this lecture, we'll be summarizing everything we learned in this section on CNN's.

2
00:00:16,770 --> 00:00:19,140
This section consisted of two main parts.

3
00:00:19,680 --> 00:00:25,080
The first part discussed how convolution works in how to incorporate it into a neural network.

4
00:00:25,890 --> 00:00:28,920
The second part looked at how to use CNN's in TensorFlow.

5
00:00:29,730 --> 00:00:35,520
You learn that CNN's are nothing but a series of convolutions, followed by pooling followed by an end.

6
00:00:36,300 --> 00:00:41,970
You'll learn that convolution acts as a pattern finder and that convolutional layers are much more efficient

7
00:00:41,970 --> 00:00:45,090
than dense layers because they use shared weights.

8
00:00:45,690 --> 00:00:47,730
This will be the case for Arden's as well.

9
00:00:49,270 --> 00:00:52,720
This may surprise you, but what you have just learned is quite powerful.

10
00:00:53,350 --> 00:00:59,260
Most courses do not look at how to use CNN's raw text, but in many cases CNN's have been shown to be

11
00:00:59,260 --> 00:01:00,910
more powerful and more performant.

12
00:01:01,180 --> 00:01:03,520
Then it ends in the field of NLP.

13
00:01:04,269 --> 00:01:10,420
In fact, historically speaking, Convolution has played a huge role in time series analysis and signal

14
00:01:10,420 --> 00:01:12,710
processing, as you recall.

15
00:01:12,730 --> 00:01:19,660
Once we convert text into vectors, what we really have is a multidimensional time series and thus convolution

16
00:01:19,960 --> 00:01:22,060
seems to be a very appropriate choice.