1
00:00:00,870 --> 00:00:09,500
We all know they learn how to design and learn and why sometimes we need two layers of perception in

2
00:00:09,510 --> 00:00:10,370
this session here.

3
00:00:10,380 --> 00:00:14,880
We have multilayer perception in this example.

4
00:00:15,220 --> 00:00:18,450
This is a word input zero input one.

5
00:00:18,450 --> 00:00:24,300
The first layer, usually in many textbook, it's called input layer.

6
00:00:24,690 --> 00:00:33,960
Then whatever the layer is between input schlager and output layer is called hidden layer, and they

7
00:00:33,960 --> 00:00:39,040
call it hidden layer because that's hidden from the output layer.

8
00:00:39,390 --> 00:00:42,070
And finally, we will have some outputs.

9
00:00:42,570 --> 00:00:45,280
Let's take a look at better example here.

10
00:00:45,780 --> 00:00:53,390
Here we have several layers and all of them are called hidden layers in Matlab as well.

11
00:00:53,670 --> 00:00:55,310
All of them actually in Matlab.

12
00:00:55,320 --> 00:00:57,270
We don't have input layer.

13
00:00:57,490 --> 00:01:01,630
I just called all of these layers hidden layer.

14
00:01:02,280 --> 00:01:05,580
But anyway, this one is a hidden layer.

15
00:01:05,580 --> 00:01:10,230
And finally we have an output layer which will give us an output.

16
00:01:11,610 --> 00:01:13,260
So let's see what we have here.

17
00:01:13,290 --> 00:01:21,920
What are our parameters, we can change the weight of each neuron as well as our bias for the system.

18
00:01:22,410 --> 00:01:28,230
So the parameters that we can change in this model are overweight.

19
00:01:31,110 --> 00:01:32,640
And the bias.

20
00:01:36,790 --> 00:01:45,610
And we need to adjust and change these parameters until we have the result, which we are looking for.

21
00:01:45,970 --> 00:01:52,890
Let's check the example of different handwriting samples we are designing and neural nets for which

22
00:01:53,380 --> 00:01:57,320
can read of our handwriting or different handwriting.

23
00:01:57,970 --> 00:02:05,980
If the system cannot, for example, read this parameter, then we need to fill it with more data.

24
00:02:05,980 --> 00:02:13,250
And also we need to adjust the parameters that we have here, including the weight as well as the bias.

25
00:02:14,050 --> 00:02:21,640
So this is an optimization agreed to and we need to optimize we need to adjust different parameters

26
00:02:21,640 --> 00:02:25,420
in order to have the best output that we are looking for.

27
00:02:26,200 --> 00:02:30,440
Adjusting different parameters is called training.

28
00:02:30,880 --> 00:02:35,140
So we need to train the system and we need to teach the system.

29
00:02:35,440 --> 00:02:42,910
The system is learning based on different parameters and based on different inputs that we are feeding

30
00:02:42,910 --> 00:02:44,800
into our network.

31
00:02:45,140 --> 00:02:48,340
This is a structure of multilayer perceptual.