1
00:00:00,760 --> 00:00:01,020
Hello.

2
00:00:01,110 --> 00:00:02,340
Welcome back.

3
00:00:02,880 --> 00:00:08,670
In convolution or neuronal networks we use polling layer to reduce the size of the inputs.

4
00:00:08,670 --> 00:00:12,040
Therefore speeding up the computation.

5
00:00:12,090 --> 00:00:14,910
There are two popular types of pooling.

6
00:00:15,360 --> 00:00:22,810
We have the Max pool and the outreach pool in taking those four by four matrix as an example.

7
00:00:23,040 --> 00:00:29,310
When we apply Max pool we get this 2 by 2 matrix shown up here.

8
00:00:29,310 --> 00:00:30,900
How did we derive this.

9
00:00:30,900 --> 00:00:32,540
This is quite simple.

10
00:00:32,610 --> 00:00:38,100
For every consecutive two by two block we take the maximum number.

11
00:00:38,100 --> 00:00:46,500
Here we have applied a filter of size 2 and a stride of 2 described in the photo size are the hyper

12
00:00:46,500 --> 00:00:55,830
parameters of the pool in there an average pooling instead of taking the maximum number we take the

13
00:00:55,950 --> 00:00:59,240
average of the numbers as we can see right here.

14
00:00:59,250 --> 00:00:59,910
Down here.