Beginning Anomaly Detection Using

IURP

NHUDV

LPSRUW

PRGHOV

OD\HUV >OD\HURXWSXW IRUOD\HULQPRGHOOD\HUV>@@

PRGHOBOD\HUV PRGHOV0RGHO LQSXWV PRGHOLQSXWRXWSXWV OD\HUV 

DFWLYDWLRQV PRGHOBOD\HUVSUHGLFW [BWUDLQ 

ILJ SOWILJXUH ILJVL]H 

SOWVXESORW 

SOWWLWOH 2ULJLQDO

SOWLPVKRZ [BWUDLQ>@UHVKDSH  FPDS JUD\

SOW[WLFNV >@

SOW\WLFNV >@

IRU

I

LQ

UDQJH  

SOWVXESORW I

SOWWLWOH &RQYROXWLRQDOOD\HUGI

OD\HUBDFWLYDWLRQ DFWLYDWLRQV>I@

SOWLPVKRZ OD\HUBDFWLYDWLRQ>@FPDS JUD\

SOW[WLFNV >@

SOW\WLFNV >@

SOWVKRZ

Figure 3-52.  Code to generate graphs of what the images look like at various 

stages of the model

Chapter 3   IntroduCtIon to deep LearnIng

---

111

As the image passes through the convolutional layers, its dimensions get reduced 

and the patterns become more apparent. While to us that might not look so much 

like a three, the model identifies those patterns from the original image and bases its 

prediction on that.

So now you have a much better understanding of what a CNN is and how Keras can 

be used to easily create and train your very own deep neural network. If you would like to 

explore the framework further, feel free to check out Appendix A. If you have any further 

questions, or would like to explore Keras beyond what’s in Appendix A, check out the 

official Keras documentation.