1
00:00:00,990 --> 00:00:09,720
You're going to change the train function manifold, the train function is live in work, marquart back

2
00:00:09,720 --> 00:00:16,800
propagation, we can change it to a different training function to know at least the training function

3
00:00:16,800 --> 00:00:17,140
here.

4
00:00:17,160 --> 00:00:25,000
It also mentioned and you can time help and train to see in a command window.

5
00:00:25,800 --> 00:00:29,150
So if I click on that, let me show it to you here.

6
00:00:30,060 --> 00:00:33,290
We have different type of training algorithm.

7
00:00:33,300 --> 00:00:41,190
If you're writing a paper, you can usually try to apply different algorithm to your data and train

8
00:00:41,190 --> 00:00:43,230
your artificial neural network.

9
00:00:43,440 --> 00:00:50,580
And then in the end, you can just analyze the effect of each training function and training algorithm

10
00:00:50,580 --> 00:00:51,410
on your system.

11
00:00:52,170 --> 00:00:53,760
So let's check it out here.

12
00:00:53,760 --> 00:01:00,930
We have a back propagation training function that to use Jacobean, which are the Bayesian regulation

13
00:01:00,930 --> 00:01:05,500
back propagation and this L.M. that is set by default.

14
00:01:06,210 --> 00:01:12,420
There are other back propagation, training function that's use gradient derivatives.

15
00:01:12,660 --> 00:01:17,100
Here we have conjugate gradient back propagation.

16
00:01:17,100 --> 00:01:22,110
We have gradient descent back propagation, which this one is a simple one.

17
00:01:22,920 --> 00:01:30,660
The only one that we have here is gradient descent with adaptive L.R. back propagation and Vyatta momentum

18
00:01:31,290 --> 00:01:37,500
so we can apply different training algorithm on our network and see the results.

19
00:01:37,520 --> 00:01:42,160
So let me just choose one of them to show you how can we see the result?

20
00:01:42,390 --> 00:01:47,720
I'm going to choose the gradient descent back propagation and applied on our network.

21
00:01:49,200 --> 00:02:00,030
So this one is train g d gradient descent back to over Advanta scrape for changing this training.

22
00:02:00,030 --> 00:02:02,130
I need to copy it first.

23
00:02:04,140 --> 00:02:10,920
I just want to put it after this feting network, after I created my new fitting network here would

24
00:02:10,920 --> 00:02:22,290
be good, just paste it and also add that that so instead of train Ilam, I'm going to choose train

25
00:02:22,290 --> 00:02:24,090
gradient descent.

26
00:02:25,560 --> 00:02:30,770
Now let's run the program to see if anything changed or not.

27
00:02:30,810 --> 00:02:33,290
Yes, it changed.

28
00:02:33,300 --> 00:02:35,710
Here we can see the training algorithm.

29
00:02:36,060 --> 00:02:38,060
Now it's a gradient descent.

30
00:02:38,610 --> 00:02:42,390
We have another factor in parameter, which was Mial.

31
00:02:42,690 --> 00:02:47,510
Now we cannot see them anymore because it was for Laurinburg algorithm.

32
00:02:48,210 --> 00:02:52,740
Let's see the feet, tools, how it was able to feed the network.

33
00:02:53,070 --> 00:02:54,320
Not a good one.

34
00:02:54,330 --> 00:03:02,550
In this example, the Lonberg algorithm or the propagation we choose a Jacob is best for this particular

35
00:03:02,550 --> 00:03:03,010
network.

36
00:03:03,270 --> 00:03:05,670
There are also some information you can read them.

37
00:03:05,670 --> 00:03:15,900
For example, if you need to use less memory, then you can use train ACG and here we have Train R,

38
00:03:15,900 --> 00:03:23,850
which takes longer but may be better for challenging problems where you just quickly reviewing each

39
00:03:23,850 --> 00:03:24,950
of the algorithms.

40
00:03:24,960 --> 00:03:28,020
There are also very good information in the help of Matlab.

41
00:03:28,320 --> 00:03:32,310
You can choose the best that feeds your neural network.