1
00:00:01,380 --> 00:00:05,340
Function activation or what we call transfer function.

2
00:00:05,580 --> 00:00:06,900
Let's take a look at this one.

3
00:00:06,930 --> 00:00:09,670
This is a simple neuron that we have here.

4
00:00:09,730 --> 00:00:15,780
There are some inputs and here we are sending them to an activation function.

5
00:00:16,110 --> 00:00:19,390
Let's talk about this activation function little bit more.

6
00:00:19,590 --> 00:00:22,250
I just got all these pictures from Matlab website.

7
00:00:22,310 --> 00:00:27,080
There are very good information and details and examples.

8
00:00:27,480 --> 00:00:30,780
So if you got time, just go there and check them out.

9
00:00:31,350 --> 00:00:32,110
Let's check it here.

10
00:00:32,880 --> 00:00:42,390
No one can use any differentiable transfer function to generate output multilayer often use log siegmund

11
00:00:42,690 --> 00:00:51,900
log sigmoid is a transfer function which generates output between zero and one as the neural net inputs

12
00:00:51,900 --> 00:00:56,390
goes from negative infinity to positive infinity.

13
00:00:56,820 --> 00:00:59,790
A famous function that we can define here.

14
00:00:59,790 --> 00:01:09,800
E F of X equals two one over one plus E to the negative X.

15
00:01:10,500 --> 00:01:15,570
This is a very famous function that we can use when designing our activation function.

16
00:01:16,020 --> 00:01:24,180
Next one that we have here is a ten sigmoid transfer function, which we show it do in Matlab with 10

17
00:01:24,200 --> 00:01:27,590
sike and then this is a sign and symbol for it.

18
00:01:28,170 --> 00:01:34,160
So it just give us the output from negative one to positive one one.

19
00:01:34,180 --> 00:01:39,100
The inputs are going from negative infinity to positive infinity.

20
00:01:39,300 --> 00:01:42,590
This is just a tangent hyperbolic.

21
00:01:42,600 --> 00:01:45,060
If you remember the function for it.

22
00:01:45,070 --> 00:01:46,950
It's let me just write it here.

23
00:01:47,110 --> 00:01:48,750
I'm going to just write it there.

24
00:01:49,860 --> 00:01:54,390
Tensioned hyperbole to show it with 10 H.

25
00:01:54,810 --> 00:02:08,580
Tenjin Hyperboles of X is equal to erase race to the X minus E to negative X over A minus the X plus

26
00:02:08,580 --> 00:02:10,080
a series of the negative X.

27
00:02:10,580 --> 00:02:15,570
But this is a very famous function that we can also use for our activation function.

28
00:02:15,600 --> 00:02:21,480
Sigmoid output neurons are often used for pattern recognition problems line here.

29
00:02:21,480 --> 00:02:27,720
We have linear output neurons that are often used for feting problems.

30
00:02:28,020 --> 00:02:36,480
And that's why the output layer, which we have a fitting problem, we use these purely or let's say

31
00:02:36,480 --> 00:02:38,030
linear transfer function.

32
00:02:38,910 --> 00:02:43,200
This is a matlab code for purely linear transfer function.

33
00:02:43,200 --> 00:02:44,670
We can just use this code.

34
00:02:45,210 --> 00:02:52,860
We often use these two in a multilayer Perceptron for our hidden layer, which we have the pattern recognition

35
00:02:52,860 --> 00:02:57,660
problem and we use this transfer function for our output layer.

36
00:02:58,050 --> 00:03:01,890
So let's check it out and see an example on Matlab.