Beginning Anomaly Detection Using

Figure 7-2.  The filter weights associated with this one-dimensional 

convolutional layer

Input:

1   0.2   0.1

Filter Weights:

Output:

10   5   15  

20   10   20

10 + 1 + 1.5

*

=

12.5

Figure 7-3.  How the first entry of the output vector is calculated using the filter 

weights. The filter weights are multiplied element-wise with the first three entries in 

the input, and the results are summed up to produce the output value

Chapter 7   temporal Convolutional networks

---

261

Input:

1   0.2   0.1

Filter Weights:

Output:

10

5   15   20

10   20

5 + 3 + 2

*

=

12.5

10

Figure 7-4.  How the second entry of the output vector is calculated using the 

filter weights. The procedure is the same as in Figure 

7-3

, but the filter weights are 

shifted right one

Input:

Filter Weights:

Output:

10 5

15   20 10

20

15 + 4 + 1

1   0.2   0.1

*

=

12.5   10

20

Figure 7-5.  How the third entry of the output vector is calculated using the filter 

weights

Input:

Filter Weights:

Output:

10 5 15

20 10 20

12.5   10   20

24

20 + 2 + 2

1   0.2   0.1

*

=

Figure 7-6.  How the last entry of the output vector is calculated using the filter 

weights

Chapter 7   temporal Convolutional networks

---

262

Now we have the output of the one-dimensional convolutional layer. These 

one-dimensional convolutional layers are quite similar to how two-dimensional 

convolutional layers work, and they comprise nearly the entirety of the two different 

TCNs we will look at: the