Beginning Anomaly Detection Using

FODVV

&11 QQ0RGXOH 

GHI

BBLQLWBB VHOI 

VXSHU &11VHOI BBLQLWBB

VHOIFRQY QQ&RQYG 

VHOIFRQY QQ&RQYG  

VHOIGHQVH QQ/LQHDU   

VHOIGHQVH  QQ/LQHDU QXPBFODVVHV

GHI

IRUZDUG VHOI[ 

[ )UHOX VHOIFRQY [

[ )UHOX VHOIFRQY [

[ )PD[BSRROG [

[ )GURSRXW [

[ [YLHZ   

[ )UHOX VHOIGHQVH [

[ )GURSRXW [

[ VHOIGHQVH [

UHWXUQ

)ORJBVRIWPD[ [GLP 

Figure 3-57.  Creation of a convolutional neural network in PyTorch

Chapter 3   IntroduCtIon to deep LearnIng

---

115

The procedure is a bit different than in Keras. In this example, the major layers were 

defined under __init__, which are your two convolutional layers and the two dense 

layers. The rest of the layers are defined under forward(). In forward(), you set x equal 

to the output of the activation function of the first convolutional layer. This new x is 

now the input of the next convolutional layer, and you set x equal to the output of the 

activation function of the second convolution layer. This same process repeats for the 

other layers, but the exact flow of data can be a bit confusing, so Figure 

3-58

 shows an 

example of what this code actually does.

The original inputs of x, self.conv1, and F.relu can be shown as such. x passes into the 

convolutional layer, and the outputs of that layer pass through the ReLU function. Then 

you get your final output X’ (Figure 

3-59

).

Figure 3-58.  F.relu is f(x), x is the training data, and self.conv1 is the first 

convolutional layer

Chapter 3   IntroduCtIon to deep LearnIng

---

116

Now, X is X’, and this new X gets passed onto the next layer (Figure 

3-60

).

Figure 3-59.  The outputs of f(x) are now the new x. Basically, x = f(x). In this case, 

the output x’ is the new x

Figure 3-60.  The new x is now the new input for the next convolutional layer

Chapter 3   IntroduCtIon to deep LearnIng

---

117

The same process repeats again, except with the new value of X (Figure 

3-61

).

And now you get the new output X

′

 (Figure 

3-62

).

This new output X

″ is then the new value of X, and the process continues.