1
00:00:00,540 --> 00:00:07,900
Optimization methods are just a different way to calculate the coefficients off of variations.

2
00:00:09,300 --> 00:00:17,040
There are two sets of optimization methods Classico optimization methods and intelligent optimization

3
00:00:17,040 --> 00:00:17,580
methods.

4
00:00:18,660 --> 00:00:28,530
Let's review some of the methods in each class and get to know some details for classical optimization

5
00:00:28,530 --> 00:00:29,000
methods.

6
00:00:29,010 --> 00:00:34,140
We have gradient descent, which we know as a short form for Ghedi.

7
00:00:34,860 --> 00:00:43,200
Gradient descent is a first order integrative optimization algorithm for finding a local minimum of

8
00:00:43,200 --> 00:00:51,480
a differentiable function in artificial neural nets, for we use gradient descent to update the coefficients

9
00:00:51,480 --> 00:00:54,130
and weights in the network.

10
00:00:55,020 --> 00:00:57,750
The other one is just adaptive learning method.

11
00:00:58,050 --> 00:01:00,870
And here we have also back propagation.

12
00:01:01,200 --> 00:01:06,600
Back propagation is short for back for propagation of errors.

13
00:01:07,410 --> 00:01:14,370
This is an algorithm for supervised learning of artificial neural networks using gradient descent.

14
00:01:15,030 --> 00:01:21,870
Given an artificial neural network and an error function, the method calculates the gradient of the

15
00:01:21,870 --> 00:01:26,550
error function with respect to the neural networks.

16
00:01:26,550 --> 00:01:36,810
Veidt, it is a generalization of the Delta rule for perceptions to multiplayer fit for neural networks.

17
00:01:37,620 --> 00:01:42,540
We are not really interested to see the details of each of these methods.

18
00:01:42,840 --> 00:01:49,800
If you want to have further more study, I can give you some reference book at the end of this course

19
00:01:49,920 --> 00:01:50,880
for your study.

20
00:01:50,880 --> 00:01:58,590
But for our use, which is practical purposes, we really don't need to know the details of how these

21
00:01:58,590 --> 00:01:59,760
methods are working.

22
00:01:59,880 --> 00:02:04,050
We only need to know when to use them and how to use them.

23
00:02:05,440 --> 00:02:13,270
Next one is intelligent optimization method, the most famous one here, we can mention E's genetic

24
00:02:13,270 --> 00:02:14,080
algorithm.

25
00:02:14,440 --> 00:02:22,390
The genetic algorithm is a method for solving both concentrated and unconcentrated optimization problems

26
00:02:22,580 --> 00:02:25,820
that is based on natural selection.

27
00:02:26,470 --> 00:02:33,270
If you are a computer engineer, you might already have some knowledge and background on genetic algorithm

28
00:02:33,580 --> 00:02:39,970
if not only learn how to use it and really no need for getting into the details.

29
00:02:41,110 --> 00:02:49,450
Swarm optimization algorithm or piso, which is a computational method that optimize a problem, buy

30
00:02:49,450 --> 00:02:57,620
into relatively trying to improve a candidate solution which record to a given measure of quality.

31
00:02:58,180 --> 00:03:05,490
And finally, we have differential evaluation algorithm and simulated annealing algorithm or assay.

32
00:03:06,070 --> 00:03:07,150
There are more.

33
00:03:07,390 --> 00:03:10,000
I just put the most famous one.

34
00:03:10,600 --> 00:03:13,950
And again, we really don't need to know the details.

35
00:03:13,960 --> 00:03:19,420
We only need to know how they are working for the practical purposes.