Beginning Anomaly Detection Using

Figure 4-38.  Plotting of accuracy shown in TensorBoard

Figure 4-39.  Plotting of loss shown in TensorBoard

Chapter 4   autoenCoders

---

152

Figure 

4-40

 shows the plotting of the accuracy of validation during the training 

process through the epochs of training.

Figure 4-40.  Plotting of validation accuracy shown in TensorBoard

Figure 4-41.  Plotting of validation loss shown in TensorBoard

Figure 

4-41

 shows the plotting of the loss of validation during the training process 

through the epochs of training.

Chapter 4   autoenCoders

---

153

 Denoising  Autoencoders

You can force the autoencoder to learn useful features by adding random noise to its 

inputs and making it recover the original noise-free data. This way the autoencoder 

can’t simply copy the input to its output because the input also contains random noise. 

The autoencoder will remove noise and produce the underlying meaningful data. 

This is called a denoising autoencoder. Figure 

4-42

 shows a depiction of a denoising 

autoencoder.

Figure 4-42.  Depiction of a denoising autoencoder

Other example is a security monitoring camera capturing some kind of hazy unclear 

picture, maybe in the dark or during adverse weather, causing a noisy image.

The logic behind the denoising autoencoder is that if you have trained your encoder 

on good normal images, and the noise, when it comes as part of the input, is not really a 

salient characteristic, it is possible to detect and remove such noise.

Figure 

4-43

 shows the basic code to import all necessary packages. Also note the 

versions of the various packages.

Chapter 4   autoenCoders

---

154

You will use the mnist images data set for this purpose. Mnist contains images for the 

digits 0 to 9 and is used for many different use cases. Figure 

4-44

 shows the code to load 

MNIST images.