1
00:00:02,040 --> 00:00:04,300
Lynn, it is a very simple architecture.

2
00:00:05,920 --> 00:00:07,550
It has three convolutional layers.

3
00:00:08,620 --> 00:00:12,700
Two of the congressional leaders also have average pulling leered.

4
00:00:13,510 --> 00:00:15,520
Notice that it is not max pooling.

5
00:00:16,060 --> 00:00:18,970
It was average pulling, as I told you earlier.

6
00:00:18,970 --> 00:00:22,600
Also, average billing was more popular in the earlier days.

7
00:00:23,330 --> 00:00:25,510
And later on, Max Pooling became more popular.

8
00:00:26,290 --> 00:00:32,640
So Limit being one of the earliest congressional neural network architectures, had average pooling.

9
00:00:34,030 --> 00:00:41,630
So the first convolutional it and in congressional it have average pulling toward converting it straight

10
00:00:41,690 --> 00:00:41,900
away.

11
00:00:41,960 --> 00:00:45,640
It gives its output to a fully connected neural network.

12
00:00:47,290 --> 00:00:53,410
If you look at the input image it took in an image of size 32, what did you do?

13
00:00:55,180 --> 00:01:02,350
And after the convolutional Lears, we had 120 such features of one by one size.

14
00:01:04,120 --> 00:01:07,480
These would fit into a fully connected neural network.

15
00:01:09,160 --> 00:01:15,350
This problem was run on amnestied only, which is handwriting recognition data.

16
00:01:16,090 --> 00:01:18,400
And it was able to achieve very good accuracy's.

17
00:01:20,050 --> 00:01:21,700
This architecture is very simple.

18
00:01:21,880 --> 00:01:25,780
In fact, I would encourage you to make this architecture in your system.

19
00:01:26,200 --> 00:01:29,450
You know, everything that you need to know to create this architecture.

20
00:01:29,710 --> 00:01:30,730
And you can run this.