Beginning Anomaly Detection Using

IURP

VNOHDUQPHWULFV

LPSRUW

URFBDXFBVFRUH

SUHGV >@

\BWUXH >@

7HVWWKHPRGHO

PRGHOHYDO 6HWPRGHOWRHYDOXDWLRQPRGH

ZLWK

WRUFKQRBJUDG 

FRUUHFW 

WRWDO 

IRU

LPDJHVODEHOV

LQ

WHVWBORDGHU

LPDJHV LPDJHVWR GHYLFH

ODEHOV ODEHOVWR GHYLFH

RXWSXWV PRGHO LPDJHV

BSUHGLFWHG WRUFKPD[ RXWSXWVGDWD

WRWDO ODEHOVVL]H 

FRUUHFW  SUHGLFWHG  ODEHOV VXP LWHP 

GHWDFKHGBSUHG SUHGLFWHGGHWDFK FSX QXPS\

GHWDFKHGBODEHO ODEHOVGHWDFK FSX QXPS\

IRU

I

LQ UDQJH

OHQ GHWDFKHGBSUHG 

SUHGVDSSHQG GHWDFKHGBSUHG>I@

\BWUXHDSSHQG GHWDFKHGBODEHO>I@

SULQW

7HVW$FFXUDF\RIWKHPRGHORQWKHWHVWLPDJHV

^` IRUPDW FRUUHFWWRWDO

SUHGV QSH\H QXPBFODVVHV >SUHGV@

\BWUXH QSH\H QXPBFODVVHV >\BWUXH@

DXF URFBDXFBVFRUH SUHGV\BWUXH

SULQW

$8&^`IRUPDW DXF

6DYHWKHPRGHOFKHFNSRLQW

WRUFKVDYH PRGHOVWDWHBGLFW  S\WRUFKBPQLVWBFQQFNSW

Figure 3-65.  Code to evaluate the model and generate the AUC score

Chapter 3   IntroduCtIon to deep LearnIng

---

121

The resulting output is shown in Figure 

3-66

.

Now you a bit more about how to create and train your own CNN in PyTorch. 

PyTorch is a bit harder to learn than Keras, which aims to make everything quite 

readable and simple, having abstracted all of the more complicated bits of code. 

TensorFlow and PyTorch are both low-level APIs that require more code to be written 

because of the lack of abstraction, but offer more flexibility in controlling exactly how 

you want everything to be. Between the two, PyTorch is easier to debug if you’re using 

the debugging tool in PyCharm. In the end, it’s all a matter of preference, although 

TensorFlow and PyTorch both perform faster on larger data sets.

Figure 3-66.  The generated accuracy score on the test set and the AUC score for 

the model

Chapter 3   IntroduCtIon to deep LearnIng

---

122

If you would like to explore PyTorch further, check out Appendix B, where we cover 

a more refined way to create models, train, and test, as well as the general functionality 

that PyTorch offers. Appendix B also applies PyTorch to the models in Chapter 

7

, which 

are done in Keras.

If you would like to learn more about PyTorch after visiting Appendix B, check out 

the official PyTorch documentation.

 Summary

In recent years, deep learning has revolutionized an incredible variety of fields. Thanks 

to deep learning, we now have self-driving cars, models that have beaten professionals in 

detecting certain cancers, instant translation between languages, etc. It is of no surprise, 

then, that deep learning has also contributed heavily to the field of anomaly detection.

In this chapter, we discussed what deep learning is and what an artificial neural 

network is. You explored two popular frameworks, Keras and PyTorch, by applying them 

to the task of image classification with the MNIST data set.

In the upcoming chapters, we will take a look at the applications to anomaly 

detection of the following types of deep learning models: