Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,300 --> 00:00:02,290
Hello and welcome back to the course and deep learning.
2
00:00:02,430 --> 00:00:07,980
Today we're talking about the neuron which is the basic building block of artificial neural networks.
3
00:00:08,010 --> 00:00:09,390
So let's get started.
4
00:00:09,390 --> 00:00:11,340
Previously we saw an image which looked like this.
5
00:00:11,340 --> 00:00:18,300
And these are actual real life neurons which have been smeared onto a gloss and color a little bit and
6
00:00:18,360 --> 00:00:19,950
they are observed through a microscope.
7
00:00:19,950 --> 00:00:22,140
So this is what they look like as you can see.
8
00:00:22,140 --> 00:00:29,730
Quite an interesting structure a body and then lots of different tails kind of branches coming out of
9
00:00:29,730 --> 00:00:30,250
them.
10
00:00:30,330 --> 00:00:32,350
And this is very interesting.
11
00:00:32,370 --> 00:00:38,400
But the question is how can we recreate that in a machine because we really need to recreate it and
12
00:00:38,400 --> 00:00:47,610
machine since the whole purpose of deep learning is to mimic how the human brain works in the hope that
13
00:00:47,820 --> 00:00:51,000
by doing so we're going to create something amazing.
14
00:00:51,000 --> 00:00:55,200
We're going to create an amazing infrastructure for machines to be able to learn.
15
00:00:55,230 --> 00:00:56,800
And why do we hope for that.
16
00:00:56,820 --> 00:01:03,480
Well because the human brain is well just happens to be one of the most powerful learning learning tools
17
00:01:03,880 --> 00:01:07,300
on the planet or like learning mechanisms on the planet.
18
00:01:07,320 --> 00:01:11,310
And we just hope that if we recreate that we'll have something as awesome as that.
19
00:01:11,310 --> 00:01:17,670
So our challenge right now our very first step to creating artificial neural networks is to recreate
20
00:01:17,700 --> 00:01:18,380
a neuron.
21
00:01:18,390 --> 00:01:19,090
So how do we do it.
22
00:01:19,110 --> 00:01:23,840
Well first of all let's have a closer look at what it actually is.
23
00:01:23,880 --> 00:01:33,180
This image was first created by a Spanish neuroscientist and Chagga Ramon Yi Kajal in 1899.
24
00:01:33,180 --> 00:01:37,780
And what he did was he died in neurons in actual brain tissue.
25
00:01:37,780 --> 00:01:39,850
And look at them under a microscope.
26
00:01:39,900 --> 00:01:43,530
And while he was looking at them he actually drew what he saw and this is what he saw.
27
00:01:43,530 --> 00:01:49,560
He saw it to your hands or two large neurons over there at the top which had all these branches coming
28
00:01:49,560 --> 00:01:57,930
off of them towards their top parts and then each one had this Araud or like thread coming out towards
29
00:01:57,930 --> 00:01:59,410
the bottom very long one.
30
00:01:59,520 --> 00:02:01,510
And so that's what he saw.
31
00:02:01,660 --> 00:02:07,800
And now you know technology has advanced quite a lot and we have seen neurons much closer in more detail
32
00:02:07,800 --> 00:02:11,890
and now we can actually draw what it looks like diagrammatic.
33
00:02:11,910 --> 00:02:13,220
So let's have a look at that.
34
00:02:13,440 --> 00:02:14,190
Here's a neuron.
35
00:02:14,190 --> 00:02:21,810
This is what it looks like very similar to what Santiago around one drew over here and here and this
36
00:02:21,810 --> 00:02:24,310
year and what we can see is that it's got a body.
37
00:02:24,570 --> 00:02:29,100
That's the main part of the neuron and then it's got some branches at the top which is called dendrites
38
00:02:29,160 --> 00:02:33,200
and it's also got an X on which is that long tail of the euro.
39
00:02:33,300 --> 00:02:38,030
And so what are these dendrites and when foreign was the axen for well.
40
00:02:38,130 --> 00:02:44,040
The key point to understand here is that neurons by themselves are pretty much useless.
41
00:02:44,040 --> 00:02:45,570
It's like it's like an ant.
42
00:02:45,600 --> 00:02:46,140
Right.
43
00:02:46,170 --> 00:02:49,640
And on its own can do my psych five ads together.
44
00:02:49,830 --> 00:02:51,170
Maybe they can pick something up.
45
00:02:51,190 --> 00:02:55,830
But again they they don't they can build an anthill or they call them establish a colony that can't
46
00:02:56,430 --> 00:02:59,340
work together as a as a huge organism.
47
00:02:59,370 --> 00:03:03,510
But at the same time when you have lots and lots of ads like you have in a million and they can build
48
00:03:03,510 --> 00:03:05,680
a whole colony they can build an anthill.
49
00:03:05,680 --> 00:03:06,600
Same thing with neurons.
50
00:03:06,600 --> 00:03:12,320
By itself it's not that strong but when you have lots of neurons together they work together to do magic.
51
00:03:12,510 --> 00:03:13,820
And how do they work together.
52
00:03:13,820 --> 00:03:14,430
That's the question.
53
00:03:14,440 --> 00:03:19,140
Well that's what the dendrites and Aksenov for so the dendrites are kind of like the receivers of the
54
00:03:19,140 --> 00:03:22,980
signal for the neuron and the axon is the transmitter of the signal for the neuron.
55
00:03:23,220 --> 00:03:26,520
And here's an image of how it all works conceptually.
56
00:03:26,520 --> 00:03:32,550
So at the top you've got on your own and you can see that it's dendrites are connected to axons of other
57
00:03:32,550 --> 00:03:35,990
neurons that are like even further away above it.
58
00:03:36,000 --> 00:03:42,930
And then the signal from your own travels down its axon and connects or passes on to the dendrites of
59
00:03:42,930 --> 00:03:44,960
the next neuron and that's how they're connected.
60
00:03:45,030 --> 00:03:53,040
And in that small image over there you can see that the axon doesn't actually touch the dendrite lot.
61
00:03:53,310 --> 00:03:59,130
A lot of machine learning or like a few machine learning scientists are very adamant about the fact
62
00:03:59,130 --> 00:04:03,650
that it doesn't touch it like the room it doesn't touch.
63
00:04:03,660 --> 00:04:06,890
It has been proven that there is no physical connection there.
64
00:04:06,960 --> 00:04:14,010
But the point that we're interested in is that that connection between them that the whole concept of
65
00:04:14,010 --> 00:04:19,590
the signal being passed that recall the sign ups you can see over there in that little image that's
66
00:04:20,300 --> 00:04:22,210
figure bracket is a sign up.
67
00:04:22,230 --> 00:04:23,820
And that's the trend we're going to be doing.
68
00:04:23,820 --> 00:04:29,820
So instead of calling our artificial neurons the lines that we're going to have or the connectors for
69
00:04:29,820 --> 00:04:34,200
artificial neurons we're now going to be calling them axons or dendrites because then the question is
70
00:04:34,200 --> 00:04:36,880
whose connection is this is it that neurons are neurons.
71
00:04:36,990 --> 00:04:39,340
We just call that good which is good to call them sign of cells.
72
00:04:39,510 --> 00:04:42,680
And that's kind of just answers all questions in a.
73
00:04:42,690 --> 00:04:47,610
That's just basically where the signal is passed doesn't matter who that element belongs to.
74
00:04:47,610 --> 00:04:51,550
They're just a representation of the signal pass and we'll see that just now.
75
00:04:51,960 --> 00:04:55,210
So basically that's how a neuron works.
76
00:04:55,210 --> 00:05:03,620
And yeah so let's move on to how are we going to represent neuron create neurons in machines are we
77
00:05:03,620 --> 00:05:09,380
moving away now we're moving away from neuroscience and moving into technology.
78
00:05:09,460 --> 00:05:10,260
And here we go.
79
00:05:10,360 --> 00:05:17,260
So here's our neuron also sometimes called the node than your own gets some input signals and it has
80
00:05:17,260 --> 00:05:18,400
an output signal.
81
00:05:18,400 --> 00:05:21,040
So dendrites and axons remember.
82
00:05:21,040 --> 00:05:27,490
But again we're going to call these sign ups and then these input signals we're going to present them
83
00:05:27,490 --> 00:05:29,050
of other neurons as well.
84
00:05:29,080 --> 00:05:35,500
So in this specific case you can see that this neuron is green you're on is getting signals from yellow
85
00:05:35,500 --> 00:05:35,860
neurons.
86
00:05:35,860 --> 00:05:41,800
And in this course we're going to try and stick to a certain color coding regime where yellow means
87
00:05:41,830 --> 00:05:42,540
an input layer.
88
00:05:42,540 --> 00:05:50,700
So basically all of the neurons that are on the outer layer on the first front of where are the signals
89
00:05:50,710 --> 00:05:52,300
coming in and by signal.
90
00:05:52,300 --> 00:05:59,200
It might be like a bit of an over overkill to call this a signal it's just basically input values so.
91
00:05:59,330 --> 00:06:04,720
So you know how even like in a simple linear regression you have input values and then you have a predicted
92
00:06:04,720 --> 00:06:05,620
value Same thing here.
93
00:06:05,620 --> 00:06:10,720
So you have input values and there they are the yellow ones and then on the right you'll see just now
94
00:06:10,720 --> 00:06:11,250
it will be red.
95
00:06:11,260 --> 00:06:12,820
It'll be the output value.
96
00:06:13,600 --> 00:06:17,140
The other thing that I wanted to point out here is that in this specific example we're looking at a
97
00:06:17,140 --> 00:06:21,320
neuron which is getting its signals from the input layer neurons.
98
00:06:21,320 --> 00:06:24,220
There are also neurons but their input their neurons.
99
00:06:24,530 --> 00:06:31,450
Sometimes you'll have neurons which get their signal from other hidden layer neurons so from other green
100
00:06:31,450 --> 00:06:35,860
neurons and the concept is going to be exactly the same I just in this case we use for simplicity's
101
00:06:35,860 --> 00:06:42,820
sake we're portraying this example and in terms of the input layer the way to think about it is in the
102
00:06:42,970 --> 00:06:49,900
in the analogy of the human brain the input layer is your senses right so whatever you can see hear
103
00:06:49,900 --> 00:06:52,290
feel touch or smell.
104
00:06:52,510 --> 00:06:57,220
And of course it's like there's there's a lot of things you can see there's a lot of information coming
105
00:06:57,220 --> 00:06:57,530
in.
106
00:06:57,730 --> 00:07:02,860
But those are your That's what your brain is limited to is pretty much a life.
107
00:07:03,010 --> 00:07:09,160
Israel lives in a box made out of bones and it's only just it's it's a mind blowing concept to think
108
00:07:09,160 --> 00:07:15,430
about that your brain is just locked in a black box and the only thing it can see can hear.
109
00:07:15,430 --> 00:07:20,710
The only thing it's getting is electrical impulses coming from these organs that you have we should
110
00:07:20,710 --> 00:07:28,210
call your ears nose eyes you know your sense of touch and we're whatever and you and your and your taste.
111
00:07:28,210 --> 00:07:34,150
Right so you're just getting signals but it basically lives in this dark black box and it's making making
112
00:07:34,150 --> 00:07:38,460
sense of the world through your senses it's phenomenal.
113
00:07:38,500 --> 00:07:38,930
And so yeah.
114
00:07:38,950 --> 00:07:43,030
So you have these inputs that are coming in in terms of human brain.
115
00:07:43,030 --> 00:07:49,540
Those are your five senses and in terms of machine learning or deep learning that is basically your
116
00:07:49,900 --> 00:07:55,500
input values are your independent variables and we'll get that in a second so your input values they
117
00:07:56,400 --> 00:08:01,450
the signal is passed through sinuses to the neuron and then your own has an output value that it passes
118
00:08:01,510 --> 00:08:03,200
further on down the chain.
119
00:08:03,550 --> 00:08:07,990
In this specific case in terms of color coding again yellow means input layer so we kind of simplifying
120
00:08:07,990 --> 00:08:11,830
everything here we're saying we're only going to have like the input layer and then we're going to have
121
00:08:11,830 --> 00:08:16,510
one hidden layer with the green which is the hinterland then we're going to have the output there right
122
00:08:16,510 --> 00:08:17,140
away.
123
00:08:17,530 --> 00:08:21,370
So just so that we can get used to these calls for now.
124
00:08:21,580 --> 00:08:24,030
So there we go that's the basic structure.
125
00:08:24,030 --> 00:08:28,390
So now let's look in a bit more detail at these different elements that we have.
126
00:08:28,390 --> 00:08:31,090
So we've got the input layer and what do we have here.
127
00:08:31,090 --> 00:08:37,090
Well we have these inputs which are in fact independent variable so depend variable one and a bit variable
128
00:08:37,090 --> 00:08:38,170
to independent variable.
129
00:08:38,170 --> 00:08:44,230
And the important thing to remember here is that these independent variables are all for one single
130
00:08:44,230 --> 00:08:44,740
observation.
131
00:08:44,740 --> 00:08:47,620
So think of it as just one row in your database.
132
00:08:47,620 --> 00:08:54,790
One observation you just take all of the independent variables you know maybe it's the age of the person
133
00:08:54,820 --> 00:09:01,270
there the amount of money in the bank account and then how how do they drive or walk to work what method
134
00:09:01,270 --> 00:09:03,070
of Shampoo's protection do they use.
135
00:09:03,080 --> 00:09:08,800
So but that's all descriptors of one specific person that you are either your training your model on
136
00:09:09,130 --> 00:09:12,510
or you're performing some prediction on.
137
00:09:12,610 --> 00:09:16,900
And the other thing you need to know about these variables is that you need to standardize them so you
138
00:09:16,900 --> 00:09:21,310
need to either standardize them which means make sure that they have a mean of zero and a variance one
139
00:09:21,340 --> 00:09:29,470
or you can also sometimes and headland will point out these traces in a bit more detail perhaps in practical
140
00:09:29,480 --> 00:09:34,090
terms you might come across these sometimes you might want to know standardising might want to normalize
141
00:09:34,090 --> 00:09:34,800
them.
142
00:09:34,990 --> 00:09:41,050
Meaning that instead of making sure the mean and very Muser and variance is one you just take you know
143
00:09:41,050 --> 00:09:46,130
to subtract the minimum value and then you divide by the maximum minus minimums by the range of your
144
00:09:46,150 --> 00:09:49,350
values and the four you get values between 0 and 1.
145
00:09:49,510 --> 00:09:53,580
And it depends on this scenario you might want to do one or the other.
146
00:09:53,590 --> 00:10:00,700
But basically you want all of these variables to be quite similar in about the same a range of values
147
00:10:00,760 --> 00:10:01,680
and why.
148
00:10:01,700 --> 00:10:02,150
Why is that.
149
00:10:02,150 --> 00:10:06,890
Well all of these values are going to go into a neural network where as we'll see just now they'll be
150
00:10:06,890 --> 00:10:12,980
added up and multiplied by weights it's added up and so on and just going to be is going to be easier
151
00:10:12,980 --> 00:10:16,990
for the neural network to process them if they're all about the same.
152
00:10:17,210 --> 00:10:23,790
And and in fact you know that's that's just how it is going to be able to work properly.
153
00:10:24,260 --> 00:10:29,210
And if you want to read more about standardization normalization and other things that you can do if
154
00:10:29,210 --> 00:10:36,440
you know what variables a good additional reading paper is called efficient back prob by young Licken
155
00:10:37,070 --> 00:10:39,630
1998 links over there.
156
00:10:39,650 --> 00:10:47,570
So we're actually going to talk more about this phenomenal person in the space of deep learning in the
157
00:10:47,680 --> 00:10:49,040
part of the course we're talking about.
158
00:10:49,040 --> 00:10:53,930
Convolutional neural networks and you'll you'll see that this is definitely a person who knows what
159
00:10:53,930 --> 00:10:55,280
he's talking about.
160
00:10:55,280 --> 00:11:00,840
He's a close friend of Jeffrey Hinton who we've already seen who very dim..
161
00:11:00,860 --> 00:11:07,070
So in this paper you'll learn more about centers ization of normalization but also you can pick up lots
162
00:11:07,070 --> 00:11:11,510
of other different tips and tricks and you'll be a good a good source for additional reading as you
163
00:11:11,510 --> 00:11:12,470
go through this course.
164
00:11:12,590 --> 00:11:15,800
So yeah check it out if you are interested in some additional reading.
165
00:11:16,290 --> 00:11:16,940
There we go.
166
00:11:16,940 --> 00:11:20,210
So that's what we do for the verbals.
167
00:11:20,390 --> 00:11:23,140
And as here we've got the output value.
168
00:11:23,180 --> 00:11:24,960
So what can our output value be.
169
00:11:25,130 --> 00:11:26,280
Well we've got a couple of options.
170
00:11:26,300 --> 00:11:32,110
Output value can be it can be continuous Like for instance price it can be binary for instance a person
171
00:11:32,110 --> 00:11:39,260
will exit or will stay or it can be categorical verbal and physical wriggler categorical verbal.
172
00:11:39,260 --> 00:11:44,060
The important thing to remember here is that in that case your output value won't be just one it'll
173
00:11:44,060 --> 00:11:50,390
be several output values because these will be a dummy variables which will be representing your categories
174
00:11:51,360 --> 00:11:57,400
and that's just this how it works and just important to remember that in that case that's how you're
175
00:11:57,410 --> 00:12:02,640
going to be getting your categories out of the artificial neural network.
176
00:12:02,810 --> 00:12:05,510
But let's go back to a simple case of one output volume.
177
00:12:05,750 --> 00:12:11,780
And now let's one more point or kind of like the point the ready made I just wanted to reiterate this
178
00:12:11,780 --> 00:12:15,360
point on the left you've got a single observation.
179
00:12:15,470 --> 00:12:20,180
So wonder if you mean from your data set and on the right you have a single direction as well and that
180
00:12:20,210 --> 00:12:21,990
is the same observation.
181
00:12:22,190 --> 00:12:28,100
So important to remember that like whatever inputs you putting in that's for one row and then the output
182
00:12:28,100 --> 00:12:29,960
you get that is for that same exact row.
183
00:12:30,080 --> 00:12:34,420
Or if you're training your neural network then you know you're putting the inputs in for that one roll
184
00:12:34,420 --> 00:12:36,440
you're putting the output in for that one row.
185
00:12:36,440 --> 00:12:42,440
So like if you want to simplify the complexity think of it as a like a simple thing a regression or
186
00:12:42,440 --> 00:12:43,990
a multivariate linear regression.
187
00:12:43,990 --> 00:12:48,150
So you're putting in your values you have the output.
188
00:12:48,180 --> 00:12:49,700
There's there's like there's no question about it.
189
00:12:49,700 --> 00:12:52,790
When we're talking about things like regression because we're so used to it.
190
00:12:52,790 --> 00:12:54,980
Same thing here it's nothing too complex.
191
00:12:54,980 --> 00:12:56,740
We're just putting in values we're getting output.
192
00:12:56,760 --> 00:13:01,310
But just remember that every time it's one row you're dealing with so you don't get confused and start
193
00:13:01,310 --> 00:13:07,940
putting in like thinking that these are different different rows that you're putting into your artificial
194
00:13:07,940 --> 00:13:12,710
neural network or something this is all just values in that one Rowse a different observation different
195
00:13:12,710 --> 00:13:18,970
characteristics or attributes relating to that one observation every single time.
196
00:13:19,160 --> 00:13:26,090
OK so next thing we want to talk about here is our sinuses is Asylum's here we've got services and they
197
00:13:26,090 --> 00:13:28,940
all actually get assigned weights weights.
198
00:13:28,940 --> 00:13:37,100
We're going to talk more about weights further down but in short weights are crucial to artificial neural
199
00:13:37,100 --> 00:13:44,040
network nerve works functioning because weights are how neural networks learn by adjusting the weights
200
00:13:44,470 --> 00:13:50,270
the neural network decides in every single case what single signal is poor and what signal is not important
201
00:13:50,270 --> 00:13:51,140
to certain neuron.
202
00:13:51,140 --> 00:13:56,300
What single gets passed along and what signal doesn't get passed along or what strength to what extent
203
00:13:56,300 --> 00:13:57,770
signals get passed along.
204
00:13:57,770 --> 00:13:59,320
So weights are crucial.
205
00:13:59,330 --> 00:14:03,590
They are and they are the things that get adjusted through the process of learning.
206
00:14:03,590 --> 00:14:07,760
Like when when you're training under artificial neural network you're basically adjusting all of the
207
00:14:07,760 --> 00:14:11,050
weights in all of the sinuses across this whole neural network.
208
00:14:11,250 --> 00:14:17,910
And that's where gradient descent and back propagation come into play and those are concepts that we'll
209
00:14:17,910 --> 00:14:19,240
also discuss.
210
00:14:19,670 --> 00:14:21,410
So basically those are the weights.
211
00:14:21,410 --> 00:14:22,940
That's what I need to know for now.
212
00:14:23,150 --> 00:14:28,420
And we've got the neurons so signals go into the neuron and what happens in Europe.
213
00:14:28,430 --> 00:14:30,520
So this is the interesting part.
214
00:14:30,530 --> 00:14:33,670
Like we're talking about the neuron today what happens inside the neuron.
215
00:14:33,890 --> 00:14:40,100
So a few things happen first thing and the first step is that all of these values that it's getting
216
00:14:40,190 --> 00:14:43,730
gets added up so it takes that added.
217
00:14:43,850 --> 00:14:51,200
So the weighted sum of all of the input values that is getting very simple it's very very straightforward
218
00:14:51,220 --> 00:14:57,200
just add up multiply by the way add them up and then it applies an activation function.
219
00:14:57,200 --> 00:15:00,960
Now we're going to talk more about activation functions the down but it's basically a function that
220
00:15:00,960 --> 00:15:09,420
is assigned to this neuron or to this whole layer and it is applied to this way to add some.
221
00:15:09,570 --> 00:15:16,890
And then from that the eurozone understands if it needs to pass on a signal if you like and that's the
222
00:15:16,890 --> 00:15:22,260
signal that it passes on that the function applied to the way that some.
223
00:15:22,260 --> 00:15:26,400
But basically depending on the function the neuron will either pass on the signal it or it won't pass
224
00:15:26,400 --> 00:15:27,640
the signal on.
225
00:15:27,870 --> 00:15:30,910
And that's exactly what happened here in step three.
226
00:15:31,430 --> 00:15:35,670
The neuron pasta's on that signal to the next neuron down the line.
227
00:15:36,090 --> 00:15:39,960
And that's what we're going to talk about in the next tutorial because it is quite an important topic.
228
00:15:39,960 --> 00:15:46,380
We want to delve deeper into the activation function but hopefully for now everything is it should be
229
00:15:46,380 --> 00:15:51,600
pretty clear how you know the input values you've got weights you've got design houses you've got something
230
00:15:51,660 --> 00:15:56,430
you know happens in the neuron you've got a way Sarmad an activation function applied and then that
231
00:15:56,430 --> 00:16:00,510
is passed down the line and that is just repeated throughout the whole neural network on and on and
232
00:16:00,510 --> 00:16:06,660
on and on you know thousands hundreds of thousands of times depending on how big how many neurons you
233
00:16:06,660 --> 00:16:09,550
have how many sign ups as you have in your neural network.
234
00:16:09,570 --> 00:16:10,110
So there we go.
235
00:16:10,110 --> 00:16:13,270
Hope you enjoyed today's tutorial code to an extent.
236
00:16:13,410 --> 00:16:15,010
Until then enjoy deep learning.
26485
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.