1
00:00:00,200 --> 00:00:02,266
Hello and welcome
back to the course on Deep Learning.

2
00:00:02,266 --> 00:00:03,833
Today we're talking about the neuron,

3
00:00:03,833 --> 00:00:07,800
which is the basic building block
of artificial neural networks.

4
00:00:07,833 --> 00:00:09,233
So let's get started.

5
00:00:09,233 --> 00:00:11,166
Previously we saw an image
which looked like this.

6
00:00:11,166 --> 00:00:14,133
And these are actual real life neurons,

7
00:00:14,133 --> 00:00:17,800
which have been smeared onto a glass and,
colored a little bit.

8
00:00:17,800 --> 00:00:19,866
And they are observed.
Through a. Microscope.

9
00:00:19,866 --> 00:00:23,333
So this is what they look like, as you can
see, quite an interesting structure.

10
00:00:23,800 --> 00:00:27,000
body and then lots of different,

11
00:00:27,000 --> 00:00:30,000
tails, kind of branches
coming out of them.

12
00:00:30,166 --> 00:00:32,266
And, this is very interesting,

13
00:00:32,266 --> 00:00:35,733
but the question is,
how can we recreate that in a machine?

14
00:00:36,000 --> 00:00:38,833
Because we really need
to recreate that machine.

15
00:00:38,833 --> 00:00:44,333
Since the whole purpose of deep learning
is to mimic how the human brain works,

16
00:00:44,933 --> 00:00:50,133
in the hope that by doing so,
we're going to create something

17
00:00:50,133 --> 00:00:53,966
amazing, we're going to create an amazing
infrastructure for machines

18
00:00:53,966 --> 00:00:55,066
to be able to learn.

19
00:00:55,066 --> 00:00:56,700
And why do we hope for that?

20
00:00:56,700 --> 00:00:58,833
Well, because the human brain is,

21
00:00:58,833 --> 00:01:02,133
well, just happens
to be one of the most powerful learning,

22
00:01:02,500 --> 00:01:07,200
learning tools on the planet
or like learning mechanisms on the planet.

23
00:01:07,200 --> 00:01:10,466
And we just hope that if we recreate that,
we'll have something.

24
00:01:10,466 --> 00:01:11,233
As awesome as that.

25
00:01:11,233 --> 00:01:12,900
So our challenge right now,

26
00:01:12,900 --> 00:01:16,866
our very first step to creating
artificial neural networks is to.

27
00:01:16,900 --> 00:01:18,233
Recreate a neuron.

28
00:01:18,233 --> 00:01:19,000
So how do we do that?

29
00:01:19,000 --> 00:01:21,300
Well,
first of all, let's have a closer look.

30
00:01:21,300 --> 00:01:23,733
At what it. Actually is.

31
00:01:23,733 --> 00:01:25,566
This image was first created.

32
00:01:25,566 --> 00:01:30,833
By, Spanish neuroscientist
Santiago Ramon Cajal, in.

33
00:01:31,566 --> 00:01:33,000
1899.

34
00:01:33,000 --> 00:01:36,066
And what he did was he died.

35
00:01:36,100 --> 00:01:37,700
Neurons in actual brain tissue.

36
00:01:37,700 --> 00:01:39,733
And look at them under a microscope.

37
00:01:39,733 --> 00:01:42,433
And while he was looking at them,
he actually drew what he saw.

38
00:01:42,433 --> 00:01:43,466
And this is what he saw.

39
00:01:43,466 --> 00:01:44,033
He saw two.

40
00:01:44,033 --> 00:01:47,033
Neurons or two
large neurons over there at the top.

41
00:01:47,100 --> 00:01:52,200
which had all of these, branches
coming out of them towards their top part.

42
00:01:52,200 --> 00:01:54,533
And then the each
one of them had this rod.

43
00:01:55,500 --> 00:01:56,300
or like thread

44
00:01:56,300 --> 00:01:59,300
coming out towards the bottom,
very long one.

45
00:01:59,400 --> 00:02:01,500
And, yeah. So that's what he saw.

46
00:02:01,500 --> 00:02:04,000
And now, you know, technology has advanced
quite a lot.

47
00:02:04,000 --> 00:02:07,733
And we, have seen neurons
much closer in more detail.

48
00:02:07,733 --> 00:02:11,133
And now we can actually draw,
what it looks like.

49
00:02:11,133 --> 00:02:13,300
Diagrammatically.
So let's have a look at that.

50
00:02:13,300 --> 00:02:15,300
Here's a neuron.
This is what it looks like.

51
00:02:15,300 --> 00:02:17,466
Very similar to. What? Santiago.

52
00:02:17,466 --> 00:02:20,400
Ramon, drew over here

53
00:02:20,400 --> 00:02:23,766
and here in this neuron,
what we can see is that it's got a body.

54
00:02:24,166 --> 00:02:26,400
Well, that's the main part of the neuron.

55
00:02:26,400 --> 00:02:29,033
And then it's got some branches at the top
which are called dendrites.

56
00:02:29,033 --> 00:02:32,566
And it's also got an axon
which is at the long tail of the neuron.

57
00:02:33,233 --> 00:02:34,300
And so what are these.

58
00:02:34,300 --> 00:02:36,866
Dendrites went for
and what's the axon for? Well.

59
00:02:38,000 --> 00:02:39,600
the key point to understand here

60
00:02:39,600 --> 00:02:43,900
is that neurons by themselves are, 
pretty much useless.

61
00:02:43,900 --> 00:02:45,866
It's like it's like an ant, right?

62
00:02:45,866 --> 00:02:49,333
An ant on its own
can't do much like five ants together.

63
00:02:49,666 --> 00:02:51,066
Maybe they can. Pick something up.

64
00:02:51,066 --> 00:02:52,933
But again, they they don't.

65
00:02:52,933 --> 00:02:55,366
They can't build an anthill
or the colony. Establish a colony.

66
00:02:55,366 --> 00:02:56,433
They can't.

67
00:02:56,433 --> 00:02:59,200
work together as a as a huge organism.

68
00:02:59,200 --> 00:03:01,466
But at the same time,
when you have lots and lots of ads,

69
00:03:01,466 --> 00:03:04,233
like you have a million ants,
they can build a whole colony.

70
00:03:04,233 --> 00:03:05,633
They can build an anthill.

71
00:03:05,633 --> 00:03:07,766
Same thing with neurons by itself. It's
not that strong.

72
00:03:07,766 --> 00:03:09,433
But when you have lots of neurons
together.

73
00:03:09,433 --> 00:03:12,366
They work together to do magic.

74
00:03:12,366 --> 00:03:14,333
And, How do they work together?
That's the question.

75
00:03:14,333 --> 00:03:16,600
Well, that's
what the. Dendrites and the axon offer.

76
00:03:16,600 --> 00:03:19,966
So the dendrites are kind of like the
receivers of the signal for the neuron.

77
00:03:19,966 --> 00:03:20,966
And the axon is the.

78
00:03:20,966 --> 00:03:23,066
Transmitter of the signal for the neuron.

79
00:03:23,066 --> 00:03:24,633
And here's an image of.

80
00:03:24,633 --> 00:03:26,433
How it all works. Conceptually.

81
00:03:26,433 --> 00:03:27,900
So at the top you've got a neuron.

82
00:03:27,900 --> 00:03:31,133
And you can see that
its dendrites are connected.

83
00:03:31,133 --> 00:03:35,433
To axons of other neurons
that are like even further away above it.

84
00:03:35,866 --> 00:03:39,733
And then the signal from this neuron
travels down its axon

85
00:03:39,900 --> 00:03:43,500
and connects or passes
on to the dendrites of the next neuron.

86
00:03:43,500 --> 00:03:44,900
And that's how they're connected.

87
00:03:44,900 --> 00:03:47,900
And, in that small image over there,
you can see that,

88
00:03:47,933 --> 00:03:52,066
the axon doesn't
actually touch the dendrite.

89
00:03:52,300 --> 00:03:56,666
A lot of, a lot of machine learning
or like a few machine learning scientists.

90
00:03:57,333 --> 00:04:00,966
are very adamant about the fact
that it doesn't touch, it's.

91
00:04:01,233 --> 00:04:03,833
Like the it doesn't touch it.

92
00:04:03,833 --> 00:04:05,533
Has been proven that there is no.

93
00:04:05,533 --> 00:04:06,833
Physical connection there.

94
00:04:06,833 --> 00:04:08,900
But the point that we're interested.

95
00:04:08,900 --> 00:04:13,933
In is that that connection between them
that, the whole concept of.

96
00:04:13,933 --> 00:04:16,200
The signal being passed,
that's. Called the sign ups.

97
00:04:16,200 --> 00:04:19,500
You can see over there
in that little image, that's,

98
00:04:20,033 --> 00:04:23,700
Finger bracket is a sign ups, and
that's the term we're going to be using.

99
00:04:23,700 --> 00:04:26,700
So instead of calling our artificial.

100
00:04:26,833 --> 00:04:27,800
Neurons

101
00:04:27,800 --> 00:04:30,400
the lines that we're going to have
or the connectors for artificial neurons,

102
00:04:30,400 --> 00:04:33,033
we're not going to be calling them
axons or dendrites.

103
00:04:33,033 --> 00:04:35,100
because then the question is
whose connection is this?

104
00:04:35,100 --> 00:04:37,600
Is it that neurons
or is this neurons we. Just call and.

105
00:04:37,600 --> 00:04:39,333
We're just going to call them synapsis.

106
00:04:39,333 --> 00:04:42,600
And that's kind of just, answers
all questions right away.

107
00:04:42,600 --> 00:04:45,033
That's just basically where the signal is
passed.

108
00:04:45,033 --> 00:04:46,133
Doesn't matter who.

109
00:04:46,133 --> 00:04:47,533
That element belongs to.

110
00:04:47,533 --> 00:04:50,166
That's just a representation
of the signal to passed.

111
00:04:50,166 --> 00:04:51,833
And we'll see that just now.

112
00:04:51,833 --> 00:04:54,833
So basically that's how a neuron works.

113
00:04:55,000 --> 00:04:57,833
And yeah. So let's move on to how are.

114
00:04:57,833 --> 00:05:03,300
We going to represent neurons or how we're
going to create neurons in, Machines.

115
00:05:03,300 --> 00:05:04,800
So we're moving away now.

116
00:05:04,800 --> 00:05:07,800
We're moving away from neuroscience
and moving into.

117
00:05:08,266 --> 00:05:09,266
technology.

118
00:05:09,266 --> 00:05:10,233
And here we go.

119
00:05:10,233 --> 00:05:13,233
So here's our neuron
also sometimes called the node.

120
00:05:13,633 --> 00:05:16,933
the neuron gets some input signals and it.

121
00:05:16,933 --> 00:05:18,233
Has an output signal.

122
00:05:18,233 --> 00:05:21,000
So dendrites and axons remember.

123
00:05:21,000 --> 00:05:23,800
But again
we're going to call these sign ups and.

124
00:05:23,800 --> 00:05:24,633
Then.

125
00:05:24,633 --> 00:05:28,933
The these input signals, we're going to
represent them with other neurons as well.

126
00:05:28,933 --> 00:05:32,866
So in this specific case
you can see that this neuron,

127
00:05:32,866 --> 00:05:35,766
this green neuron is getting signals
from yellow neurons.

128
00:05:35,766 --> 00:05:37,400
And in this course we're going to try.

129
00:05:37,400 --> 00:05:39,133
And stick to, a.

130
00:05:39,133 --> 00:05:42,466
Certain color coding regime
where yellow means an input layer.

131
00:05:42,466 --> 00:05:44,966
So basically all of the neurons.

132
00:05:44,966 --> 00:05:48,766
That are on the outer layer
or in the first front of,

133
00:05:48,800 --> 00:05:51,466
where the signal signals coming in, and.

134
00:05:51,466 --> 00:05:57,033
By signal it might be like a bit
of an over overkill to call this a signal.

135
00:05:57,033 --> 00:05:58,700
It's just basically input values.

136
00:05:58,700 --> 00:06:01,933
So so you know how even like
in a simple linear regression

137
00:06:01,933 --> 00:06:04,966
you have input values
and then you have a a predicted value.

138
00:06:04,966 --> 00:06:05,533
Same thing here.

139
00:06:05,533 --> 00:06:07,100
So you have input values.

140
00:06:07,100 --> 00:06:08,900
And there they are the yellow ones.

141
00:06:08,900 --> 00:06:11,166
And then on the right
we will see just now it'll be red.

142
00:06:11,166 --> 00:06:12,166
It'll be the output value.

143
00:06:13,466 --> 00:06:15,366
the, the
thing that I wanted to point out here

144
00:06:15,366 --> 00:06:17,466
is that in this specific example,
we're looking at a neuron

145
00:06:17,466 --> 00:06:21,166
which is getting its signals
from the input layer neurons.

146
00:06:21,166 --> 00:06:22,100
So they're also neurons.

147
00:06:22,100 --> 00:06:24,466
But they're they're input layer neurons.

148
00:06:24,466 --> 00:06:26,600
sometimes you'll have neurons which

149
00:06:26,600 --> 00:06:30,366
get their signal
from other hidden layer neurons.

150
00:06:30,366 --> 00:06:31,666
So from other. Neurons.

151
00:06:31,666 --> 00:06:33,966
And the concept is going to be exactly
the same in just in this case.

152
00:06:33,966 --> 00:06:36,966
We for simplicity's sake,
we're portraying this example.

153
00:06:37,500 --> 00:06:40,500
And in terms of the input layer,
the way to think about it is,

154
00:06:40,566 --> 00:06:44,566
in, in the analogy of the human brain,

155
00:06:45,200 --> 00:06:48,000
the input layer is your senses, right?

156
00:06:48,000 --> 00:06:51,000
So whatever you can see here feel.

157
00:06:51,033 --> 00:06:52,366
touch or smell.

158
00:06:52,366 --> 00:06:55,766
And of course, it's like there's,
there's a lot of things you can see.

159
00:06:55,766 --> 00:06:57,600
There's a lot of information coming in.

160
00:06:57,600 --> 00:06:59,600
But those are your
that's what your brain is.

161
00:06:59,600 --> 00:07:02,600
Limited to is pretty much a, like.

162
00:07:02,866 --> 00:07:05,966
It's pretty much lives in a box
made out of bones.

163
00:07:05,966 --> 00:07:09,800
And it's only just, it's it's
a mind blowing concept to think about that

164
00:07:09,900 --> 00:07:14,700
your brain is just locked in a black box,
and the only thing like it can't see,

165
00:07:14,700 --> 00:07:15,333
you can't hear.

166
00:07:15,333 --> 00:07:17,266
The only thing it's getting
is electrical impulses

167
00:07:17,266 --> 00:07:21,300
coming from these, organs that you have,
which are called your ears.

168
00:07:21,300 --> 00:07:24,900
And those eyes,
you know, your sense of touch and.

169
00:07:26,166 --> 00:07:28,133
Whatever, and you and your and your taste.

170
00:07:28,133 --> 00:07:28,266
Right?

171
00:07:28,266 --> 00:07:31,433
So it's just getting signals,
but it basically lives in this dark

172
00:07:31,433 --> 00:07:35,966
black box, and it's making, making
sense of the world through your senses.

173
00:07:35,966 --> 00:07:37,866
It's, it's it's phenomenal.

174
00:07:37,866 --> 00:07:38,900
and so, yeah.

175
00:07:38,900 --> 00:07:42,933
So you have these, inputs that are
coming in in terms of human brain.

176
00:07:42,933 --> 00:07:43,933
Those are your five senses.

177
00:07:43,933 --> 00:07:46,200
And, in terms of machine learning.

178
00:07:46,200 --> 00:07:49,433
Or deep learning, that is basically your.

179
00:07:49,833 --> 00:07:50,500
input values.

180
00:07:50,500 --> 00:07:52,766
So your independent variables
and we'll get that in a second.

181
00:07:52,766 --> 00:07:57,933
So, your input values, they 
the signal is passed through synapsis.

182
00:07:57,933 --> 00:07:59,466
To your neuron, and then your neuron.

183
00:07:59,466 --> 00:08:02,600
Has an output value that it passes
it further on down the chain.

184
00:08:03,400 --> 00:08:05,166
In this specific case
in terms of color coding.

185
00:08:05,166 --> 00:08:06,900
Again yellow means input layer.

186
00:08:06,900 --> 00:08:08,500
So we kind of simplifying everything here.

187
00:08:08,500 --> 00:08:11,066
We're saying we're only going
to have like the input layer.

188
00:08:11,066 --> 00:08:12,300
Then we're going to have one.

189
00:08:12,300 --> 00:08:14,966
Hidden. Layer with the green
which is a hidden layer.

190
00:08:14,966 --> 00:08:17,400
And then we're going to have the output
layer right away.

191
00:08:17,400 --> 00:08:20,100
So just so that we can get used
to these colors for now.

192
00:08:20,100 --> 00:08:22,066
So there we go.

193
00:08:22,066 --> 00:08:23,933
That's the basic structure.

194
00:08:23,933 --> 00:08:25,200
So now let's look in a bit more.

195
00:08:25,200 --> 00:08:28,300
Detail
at these different elements that we have.

196
00:08:28,300 --> 00:08:29,833
So we've got the input layer.

197
00:08:29,833 --> 00:08:31,000
And what do we have here.

198
00:08:31,000 --> 00:08:35,066
Well we have These inputs
which are in fact independent

199
00:08:35,066 --> 00:08:37,200
variable soon depend variable one
independent variable two.

200
00:08:37,200 --> 00:08:38,566
Independent variable m.

201
00:08:38,566 --> 00:08:41,666
The important thing to remember here
is that these

202
00:08:41,666 --> 00:08:44,666
independent variables
are all for one single observation.

203
00:08:44,666 --> 00:08:47,500
So think of it as just one
row in your database.

204
00:08:47,500 --> 00:08:48,900
One observation.

205
00:08:48,900 --> 00:08:51,900
You just take all of the independent
variables.

206
00:08:51,900 --> 00:08:55,166
you know, maybe it's, 
the age of the person there,

207
00:08:55,366 --> 00:08:59,633
the amount of money in the bank accounts,
and then how how do they drive

208
00:08:59,633 --> 00:09:00,700
or walk to work?

209
00:09:00,700 --> 00:09:02,933
What method of transportation do they use?

210
00:09:02,933 --> 00:09:05,466
So but that's all descriptors
of one specific.

211
00:09:05,466 --> 00:09:07,100
Person that you are.

212
00:09:07,100 --> 00:09:09,200
Either you're training. Your model on.

213
00:09:09,200 --> 00:09:12,900
or you're performing some prediction
on, and.

214
00:09:13,100 --> 00:09:13,866
the other thing you need to.

215
00:09:13,866 --> 00:09:16,233
Know about these variables
is that you need to standardize them.

216
00:09:16,233 --> 00:09:18,800
So you need to either standardize them,
which means,

217
00:09:18,800 --> 00:09:21,266
make sure that they have a mean of zero
and variance one.

218
00:09:21,266 --> 00:09:23,400
Or you. Can also. Sometimes.

219
00:09:23,400 --> 00:09:26,600
And Hedlund will point out
these situations in a bit more.

220
00:09:27,333 --> 00:09:29,733
detail.
Perhaps in the practical tutorials.

221
00:09:29,733 --> 00:09:30,800
You might come across these.

222
00:09:30,800 --> 00:09:32,233
Sometimes you might want to

223
00:09:32,233 --> 00:09:36,066
not standardize, you might want to
normalize them, meaning that instead of,

224
00:09:36,600 --> 00:09:39,833
making sure the mean and I mean the zero
variance is one you just take,

225
00:09:39,900 --> 00:09:43,800
you know, to subtract the minimum value,
and then you divide by the maximum

226
00:09:43,800 --> 00:09:44,433
minus minimum.

227
00:09:44,433 --> 00:09:48,933
So by the range of your values and the for
you get values between 0 and 1.

228
00:09:49,400 --> 00:09:52,333
And it depends on the scenario.

229
00:09:52,333 --> 00:09:53,500
You might want to do one or the other.

230
00:09:53,500 --> 00:09:57,600
But basically you want all of these
variables to be quite similar in the, in

231
00:09:57,766 --> 00:09:58,733
about. The same.

232
00:09:59,700 --> 00:10:00,600
range of values.

233
00:10:00,600 --> 00:10:03,200
And why, why is that? Well.

234
00:10:03,200 --> 00:10:03,466
all of.

235
00:10:03,466 --> 00:10:04,033
These values are going.

236
00:10:04,033 --> 00:10:06,433
To go into a neural network
where as we'll see just now,

237
00:10:06,433 --> 00:10:10,300
they'll be added up and multiplied
by weights added up and so on.

238
00:10:10,300 --> 00:10:13,033
And just going to be
this is going to be easier for.

239
00:10:13,033 --> 00:10:16,566
The neural network to process them,
if they're all about the same.

240
00:10:17,100 --> 00:10:19,933
And and in fact, you know, that's, that's

241
00:10:19,933 --> 00:10:23,300
just how it is
going to be able to work properly.

242
00:10:24,100 --> 00:10:25,200
And if you want to read.

243
00:10:25,200 --> 00:10:28,333
More about,
standardization, normalization and other.

244
00:10:28,366 --> 00:10:29,266
Things that. You can do if you know.

245
00:10:29,266 --> 00:10:32,733
What variables,
a good additional reading.

246
00:10:33,666 --> 00:10:37,766
paper is called Efficient Backprop
by Yann LeCun, 1998.

247
00:10:38,433 --> 00:10:39,400
the links over there.

248
00:10:39,400 --> 00:10:41,333
So Yann LeCun, we're actually going.

249
00:10:41,333 --> 00:10:45,466
To talk more about, this,
phenomenal person in the space of deep

250
00:10:45,466 --> 00:10:49,500
learning in, the part of the course
where we're talking about convolutional.

251
00:10:49,500 --> 00:10:50,300
Neural. Networks.

252
00:10:50,300 --> 00:10:53,166
And you'll, you'll see that
this is definitely a person

253
00:10:53,166 --> 00:10:55,166
who knows what he's talking about.

254
00:10:55,166 --> 00:10:58,800
He's a close friend of,
Geoffrey Hinton, who we've already seen,

255
00:10:59,333 --> 00:11:02,266
who've already mentioned. So, in.

256
00:11:02,266 --> 00:11:03,966
This paper, you'll learn more about.

257
00:11:03,966 --> 00:11:06,333
Standardization and. Normalization,
but also you can.

258
00:11:06,333 --> 00:11:09,466
Pick up lots of other different
tips and tricks, and it'll be a good

259
00:11:09,466 --> 00:11:12,066
a good source for additional reading
as you go through this course.

260
00:11:12,066 --> 00:11:13,566
So definitely check it out.

261
00:11:13,566 --> 00:11:17,033
if you're interested in

262
00:11:18,900 --> 00:11:21,900
so yeah, check it out if you're interested
in some additional reading.

263
00:11:21,900 --> 00:11:24,500
There we go. So that's.

264
00:11:24,500 --> 00:11:26,700
What we need to do with the. Variables.

265
00:11:26,700 --> 00:11:29,500
And here we've got. The output value.

266
00:11:29,500 --> 00:11:31,500
So what can our output value be.

267
00:11:31,500 --> 00:11:34,500
Well we got a couple options.

268
00:11:34,566 --> 00:11:35,800
well we've got a couple of options.

269
00:11:35,800 --> 00:11:38,433
Output value can be it can be continuous.

270
00:11:38,433 --> 00:11:39,666
Like for instance price.

271
00:11:39,666 --> 00:11:40,833
It can be binary.

272
00:11:40,833 --> 00:11:43,800
For instance a person will exit
or. Will stay or it can.

273
00:11:43,800 --> 00:11:45,700
Be a categorical variable.

274
00:11:45,700 --> 00:11:48,700
And If it's a kind of
categorical variable.

275
00:11:48,800 --> 00:11:52,466
The important thing to remember here
is that in that case, your output value.

276
00:11:52,533 --> 00:11:53,866
Won't be just one. It'll be.

277
00:11:53,866 --> 00:11:55,500
Several output values.

278
00:11:55,500 --> 00:11:57,466
Because these will be your dummy
variables.

279
00:11:57,466 --> 00:11:59,933
Which will be
representing your categories.

280
00:12:00,900 --> 00:12:02,700
And that just this how it works.

281
00:12:02,700 --> 00:12:07,133
And it's just important to remember that
in that case, that's how you're going

282
00:12:07,133 --> 00:12:11,000
to be getting your categories out of the, 
artificial neural network.

283
00:12:11,766 --> 00:12:14,833
but let's go back to a simple case
of one output value.

284
00:12:15,233 --> 00:12:18,233
And now let's, one more point
or kind of like an,

285
00:12:18,600 --> 00:12:19,700
the point that we've already made.

286
00:12:19,700 --> 00:12:22,600
I just wanted. To reiterate this point.

287
00:12:22,600 --> 00:12:24,966
on the left,
you've got a single observation.

288
00:12:24,966 --> 00:12:27,266
So one of your from your data set.

289
00:12:27,266 --> 00:12:29,233
And on the right
you have a single observation as well.

290
00:12:29,233 --> 00:12:31,666
And that is the same observation.

291
00:12:31,666 --> 00:12:33,833
So important to remember that

292
00:12:33,833 --> 00:12:37,000
like whatever inputs you're putting in
that's for one row.

293
00:12:37,000 --> 00:12:39,600
And then the output you get
that is for that same exact row.

294
00:12:39,600 --> 00:12:41,100
Or if you're training your,

295
00:12:41,100 --> 00:12:43,766
neural network, then, you know,
you're putting the inputs in for that one

296
00:12:43,766 --> 00:12:45,900
row, you're putting the output in
for that one row.

297
00:12:45,900 --> 00:12:49,700
So like if you want to
simplify the complexity, think of it as a.

298
00:12:50,333 --> 00:12:53,533
like a simple linear regression
or a multivariate or linear regression.

299
00:12:53,533 --> 00:12:56,233
So you're putting in your. values.

300
00:12:56,233 --> 00:12:57,633
You have your output.

301
00:12:57,633 --> 00:13:00,466
There's, there's like there's no question
about it when we're talking about,

302
00:13:00,466 --> 00:13:02,333
things like. Regression
because we're so used to it.

303
00:13:02,333 --> 00:13:04,500
Same thing
here. It's it's nothing too complex.

304
00:13:04,500 --> 00:13:06,300
We're just putting in values.
We're getting output.

305
00:13:06,300 --> 00:13:07,333
But just remember.

306
00:13:07,333 --> 00:13:09,300
That every time
it's one row you're dealing with so you.

307
00:13:09,300 --> 00:13:10,500
Don't get confused.

308
00:13:10,500 --> 00:13:14,900
And start putting in like, thinking
that these are different, different rows

309
00:13:14,900 --> 00:13:18,666
that you're putting into your, 
artificial neural network or something.

310
00:13:18,666 --> 00:13:20,800
This is all just values in that one row.

311
00:13:20,800 --> 00:13:24,400
So different observation,
different characteristics of or attributes

312
00:13:24,400 --> 00:13:27,400
relating to that one observation
every single time.

313
00:13:27,966 --> 00:13:29,000
Okay.

314
00:13:29,000 --> 00:13:31,300
So next thing
that we want to talk about here is.

315
00:13:31,300 --> 00:13:34,000
or the sign ups is, is a sign of sense.

316
00:13:34,000 --> 00:13:37,366
here we've got sentences and they all
actually get assigned weights.

317
00:13:38,066 --> 00:13:38,466
weights.

318
00:13:38,466 --> 00:13:41,333
We're going to talk more about weights
for the down.

319
00:13:41,333 --> 00:13:46,466
But in short, weights are crucial
to artificial.

320
00:13:46,466 --> 00:13:48,600
Neural networks functioning because.

321
00:13:48,600 --> 00:13:51,300
Weights are how neural networks learn.

322
00:13:51,300 --> 00:13:54,866
By adjusting the weights,
the neural network

323
00:13:54,866 --> 00:13:58,666
decides in every single case
what single signal is important.

324
00:13:58,666 --> 00:14:00,900
What signal is not.
Important to a certain neuron, what.

325
00:14:00,900 --> 00:14:02,566
Signal gets passed along and what signal.

326
00:14:02,566 --> 00:14:05,100
Doesn't get passed along,
or to what strength?

327
00:14:05,100 --> 00:14:07,233
To what extent signals get passed along.

328
00:14:07,233 --> 00:14:08,833
So weights are crucial.

329
00:14:08,833 --> 00:14:12,000
They are,
and they are the things that get adjusted.

330
00:14:12,000 --> 00:14:13,633
Through the process of learning.
Like when.

331
00:14:13,633 --> 00:14:15,133
When you're training on artificial.

332
00:14:15,133 --> 00:14:18,033
Neural network, you're basically.
Adjusting all of the weights.

333
00:14:18,033 --> 00:14:20,400
In all of the sign ups across
this whole neural network.

334
00:14:20,400 --> 00:14:23,400
And and that's where gradient descent and,

335
00:14:23,800 --> 00:14:26,200
backpropagation come into play.

336
00:14:26,200 --> 00:14:28,666
And those are concepts
that we'll also discuss.

337
00:14:28,666 --> 00:14:30,900
so. Basically those are the weights.

338
00:14:30,900 --> 00:14:32,666
That's all we. Need to know for now.

339
00:14:32,666 --> 00:14:34,166
And here we've got the neuron.

340
00:14:34,166 --> 00:14:37,666
So signals go into the neuron
and what happens in the neuron.

341
00:14:37,933 --> 00:14:40,100
So this is the interesting part.

342
00:14:40,100 --> 00:14:41,700
Like we're talking about the neuron today.

343
00:14:41,700 --> 00:14:43,366
What happens inside the neuron.

344
00:14:43,366 --> 00:14:45,033
So a few things happen.

345
00:14:45,033 --> 00:14:49,666
First thing and the first step is that
all of these values that it's getting.

346
00:14:49,666 --> 00:14:50,900
Gets added up.

347
00:14:50,900 --> 00:14:55,733
So it Takes the added
so the weighted sum of all of.

348
00:14:55,733 --> 00:14:59,066
The input values
that it's getting very simple.

349
00:14:59,066 --> 00:14:59,400
Right.

350
00:14:59,400 --> 00:15:00,733
It's very, very straightforward.

351
00:15:00,733 --> 00:15:03,500
Just add up multiply
by the way. Add them up.

352
00:15:03,500 --> 00:15:06,500
and then it applies an activation
function.

353
00:15:06,700 --> 00:15:09,433
Now we're going to talk more
about activation functions further down.

354
00:15:09,433 --> 00:15:10,400
But it's basically a function

355
00:15:10,400 --> 00:15:13,400
that is assigned to this neuron
or to this whole layer.

356
00:15:13,766 --> 00:15:16,000
And it.

357
00:15:16,000 --> 00:15:19,000
Is applied to this weighted. Sum.

358
00:15:19,066 --> 00:15:24,166
And then from that the neuron understands,
if it needs to pass.

359
00:15:24,166 --> 00:15:25,833
On a signal like.

360
00:15:25,833 --> 00:15:28,800
The, that's the signal
that it passes on that, the function

361
00:15:28,800 --> 00:15:31,800
applied to the weighted sum.

362
00:15:31,800 --> 00:15:33,466
But basically depending on the function,

363
00:15:33,466 --> 00:15:36,633
the neuron will either pass on a signal it
or it won't pass this signal on.

364
00:15:37,366 --> 00:15:40,133
And that's exactly what happened here
in step three.

365
00:15:40,133 --> 00:15:44,766
the neuron passes on that signal
to the next neuron down the line.

366
00:15:45,600 --> 00:15:45,933
and that's.

367
00:15:45,933 --> 00:15:48,300
What we're going to talk about
in the next tutorial, because it is.

368
00:15:48,300 --> 00:15:53,533
Quite an important topic we want to delve
deeper into, the activation function.

369
00:15:53,533 --> 00:15:57,333
But hopefully for now everything is
it should be pretty clear how, you know,

370
00:15:57,333 --> 00:15:58,633
you've got input values, you've got.

371
00:15:58,633 --> 00:16:01,966
Weights, you've got the synopses, you've
got something that you know, happens in

372
00:16:01,966 --> 00:16:05,233
the neuron, you've got the weighted sum
and then the activation function applied.

373
00:16:05,233 --> 00:16:06,700
And then that is paused down the line.

374
00:16:06,700 --> 00:16:07,766
And that is just repeated

375
00:16:07,766 --> 00:16:11,000
throughout the whole neural network
on and on and on and on.

376
00:16:11,000 --> 00:16:14,100
you know,
thousands, hundreds of thousands of times

377
00:16:14,100 --> 00:16:16,400
depending on how big,
how many neurons you have,

378
00:16:16,400 --> 00:16:19,033
how many sign ups as you have
in your neural network.

379
00:16:19,033 --> 00:16:19,666
So there we go.

380
00:16:19,666 --> 00:16:21,600
Hope you enjoyed today's tutorial.

381
00:16:21,600 --> 00:16:22,766
Can't wait to see you next time.

382
00:16:22,766 --> 00:16:24,600
And until then, enjoy deep learning!