Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,120 --> 00:00:03,750
In this section, we will go through the fundamental concepts of deep learning.
2
00:00:05,130 --> 00:00:09,690
To understand deep learning, we have to know its position among similar or related fields.
3
00:00:10,380 --> 00:00:13,110
Artificial intelligence is the biggest field among all.
4
00:00:13,950 --> 00:00:18,120
The term artificial intelligence is made up of the words artificial and intelligence.
5
00:00:18,900 --> 00:00:23,820
Artificial refers to human creations, and intelligence refers to the capacity for understanding.
6
00:00:24,600 --> 00:00:27,900
AI enables machines to comprehend similarly to humans.
7
00:00:28,650 --> 00:00:31,590
Machine learning is a subset of artificial intelligence.
8
00:00:32,250 --> 00:00:35,040
One of technique in machine learning is neural networks.
9
00:00:35,640 --> 00:00:40,200
Deep learning is a kind of neural networks which has many distinct layers and give a more sophisticated
10
00:00:40,200 --> 00:00:42,240
results than simple neural networks.
11
00:00:43,310 --> 00:00:46,760
Now we are going to review a fundamental concept of neural networks.
12
00:00:47,770 --> 00:00:52,270
It is well known that neural networks can solve challenging problems by mimicking the behavior of the
13
00:00:52,270 --> 00:00:53,050
human brain.
14
00:00:53,860 --> 00:00:58,390
The input data is processed through several stack layers of artificial neurons to create the desired
15
00:00:58,390 --> 00:00:58,780
output.
16
00:01:00,050 --> 00:01:01,550
What is a neuron anyway?
17
00:01:02,270 --> 00:01:07,490
The neuron, also known as a nodal unit, is the basic unit of computation in a neural networks.
18
00:01:08,210 --> 00:01:13,430
After receiving input from another node or from an external source, the note processes the input and
19
00:01:13,430 --> 00:01:14,510
produces output.
20
00:01:15,290 --> 00:01:16,970
Its input has an associated weight.
21
00:01:16,970 --> 00:01:19,130
W the function F is applied to the weight.
22
00:01:19,130 --> 00:01:19,700
It's somebody.
23
00:01:19,700 --> 00:01:20,150
No.
24
00:01:20,990 --> 00:01:23,330
Function F is called activation function.
25
00:01:24,070 --> 00:01:28,930
This whole unit is called the perceptron, the simplest neural network architecture.
26
00:01:29,700 --> 00:01:32,220
Here are the various components of a perceptron.
27
00:01:32,640 --> 00:01:35,520
Input is the group of features that the model used to learn.
28
00:01:36,120 --> 00:01:40,830
For instance, an array of pixel values from an image can be the input for object detection.
29
00:01:41,580 --> 00:01:46,440
With its primary function is to prioritize those features that contribute the most to learning.
30
00:01:47,590 --> 00:01:48,240
Bias.
31
00:01:48,250 --> 00:01:53,440
Its role is to see the value produced by the activation function to the left or right to fit the prediction
32
00:01:53,440 --> 00:01:54,430
with the data of better.
33
00:01:55,270 --> 00:01:59,650
The activation function's purpose is to introduce nonlinearity into the neurons output.
34
00:02:00,220 --> 00:02:05,230
This is crucial as a lot of real world data is non-linear, and we want the neural networks to learn
35
00:02:05,230 --> 00:02:06,370
this nonlinear data.
36
00:02:07,500 --> 00:02:10,530
The perceptron represents how a single neuron performs.
37
00:02:11,160 --> 00:02:12,990
What if we take a lot of perceptron?
38
00:02:14,010 --> 00:02:18,210
We will have a multi layer networks with the input data pass in the forward direction.
39
00:02:19,020 --> 00:02:23,010
The layer of the feedforward neural networks contains several neurons or nodes.
40
00:02:23,650 --> 00:02:27,600
Notes in adjacent layers have connections, and each connection has a weight w.
41
00:02:28,410 --> 00:02:31,680
Feedforward networks can consist of three types of nodes.
42
00:02:32,370 --> 00:02:35,760
Input nodes bring data from the outside world into the networks.
43
00:02:36,420 --> 00:02:38,820
No computation is performed at the input node.
44
00:02:38,820 --> 00:02:41,100
Only information is passed to the hidden node.
45
00:02:41,900 --> 00:02:45,710
He did not computes and transfers data from the input not to the output node.
46
00:02:46,560 --> 00:02:50,550
Computation will be performed and networks output will be generated to put No.
47
00:02:51,370 --> 00:02:56,410
In general, deep learning networks have numerous hidden layers which make them called deep neural networks.
48
00:02:57,040 --> 00:03:01,690
We could not find any solid reference on how many hidden layers that make a neural networks called deep
49
00:03:01,690 --> 00:03:02,800
neural networks.
50
00:03:03,460 --> 00:03:07,990
However, a deep learning hidden measures are associated with extracting features, whereas standard
51
00:03:07,990 --> 00:03:12,670
neural networks use neurons to transmit input to get output with the help of various connections.
52
00:03:14,010 --> 00:03:18,180
Now we are going to talk about why the residual network or Internet is very important.
53
00:03:18,980 --> 00:03:22,790
More hidden layers typically give networks the chance to learn more effectively.
54
00:03:23,480 --> 00:03:26,960
However, it could lead to gradient problems like fencing or exploding.
55
00:03:27,820 --> 00:03:31,540
The product of any derivatives will occur in a network within hidden layers.
56
00:03:32,200 --> 00:03:37,180
The gradient will decrease exponentially if the derivatives are small, which causes the vanishing gradient
57
00:03:37,180 --> 00:03:37,870
problem.
58
00:03:38,700 --> 00:03:43,740
However, if the derivatives are significant, the gradient will grow exponentially, creating the exploding
59
00:03:43,740 --> 00:03:44,760
gradient problem.
60
00:03:45,530 --> 00:03:50,060
If anything gradient occurs, the gradient may be zero and the network training will stop.
61
00:03:50,940 --> 00:03:56,070
The moderates may grow very large if an exploding gradient occurs leading to overflow or nan, which
62
00:03:56,070 --> 00:03:58,470
prevents the weight from being updated any longer.
63
00:03:59,250 --> 00:04:00,180
For those issues.
64
00:04:00,180 --> 00:04:02,310
Rest nets offer an alternative solution.
65
00:04:03,090 --> 00:04:08,180
The fundamental principle of risk net is to copy the prior result x and end to the subsequent result
66
00:04:08,190 --> 00:04:08,940
f x.
67
00:04:10,000 --> 00:04:13,060
The previous slides explain how perceptron performs.
68
00:04:13,720 --> 00:04:18,790
There is a function F in every perceptron, which is a nonlinear function known as an activation function.
69
00:04:19,560 --> 00:04:22,290
So what is exactly an activation function?
70
00:04:23,710 --> 00:04:29,020
An activation function is inspired by brain activities in which different neurons are activated by different
71
00:04:29,020 --> 00:04:29,860
stimulus.
72
00:04:30,460 --> 00:04:35,170
For example, when we touch file, certain neurons are activated, causing us to believe that we have
73
00:04:35,170 --> 00:04:38,680
thus far feel the pain of the heat and immediately remove our hand.
74
00:04:39,590 --> 00:04:44,150
Inactivation function in neural networks makes the choice of whether to activate a neuron or not.
75
00:04:45,080 --> 00:04:49,820
Therefore, the activation function is a straightforward mathematical calculation to decide whether
76
00:04:49,820 --> 00:04:53,570
or not the input from the neuron is significant during the prediction process.
77
00:04:54,290 --> 00:04:59,510
Aside from that, the purpose of an activation function is to add nonlinearity to the neural networks
78
00:05:00,140 --> 00:05:05,000
without activation functions and neural networks is simply a linear regression model that can struggle
79
00:05:05,000 --> 00:05:06,740
to learn any complex task.
80
00:05:07,900 --> 00:05:10,390
There are several activation functions available.
81
00:05:11,170 --> 00:05:15,970
The sigmoid activation function will take a real valued input and put it between zero and one.
82
00:05:17,050 --> 00:05:21,910
The hyperbolic tangent activation function will take a real valued input and push the value into the
83
00:05:21,910 --> 00:05:23,830
range of values minus one and one.
84
00:05:24,960 --> 00:05:27,310
Rectified linear unit, which is abbreviated.
85
00:05:27,310 --> 00:05:30,750
This review is one of the most popular activation functions.
86
00:05:31,410 --> 00:05:33,420
It's a very straightforward formula.
87
00:05:34,170 --> 00:05:36,750
But both accuracy and speed are performed well.
88
00:05:37,520 --> 00:05:40,400
This activation function only limits the number of zero.
89
00:05:41,210 --> 00:05:46,580
This means that if X is less than zero, then if x equals zero and if x is greater than zero, then
90
00:05:46,580 --> 00:05:47,900
if x equals x.
91
00:05:48,930 --> 00:05:53,700
Miki will modifies review by allowing small negative values when values are less than zero.
92
00:05:54,240 --> 00:05:56,370
Lakeville is used in all of three.
93
00:05:57,510 --> 00:06:02,660
This is a self regularized non monotonic activation function when compared to review.
94
00:06:02,700 --> 00:06:06,960
The missed function consistently improves the accuracy of the neural networks architecture.
95
00:06:07,880 --> 00:06:10,610
This activation function is used in all of four.
96
00:06:11,760 --> 00:06:14,190
Serial stands for sigmoid Weighted linear unit.
97
00:06:14,790 --> 00:06:19,560
This activation function is calculated by multiplying the sigmoid function by its input.
98
00:06:20,100 --> 00:06:21,570
Cielo is used in your office.
99
00:06:21,570 --> 00:06:22,140
Seven.
100
00:06:22,830 --> 00:06:24,290
See you in the next video.
10205
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.