Summary

Congratulations! You have developed your first end-to-end image recognition AI service.

We also learned how to create your Python ML development environment and install and manage your dependencies using pip and virtual environments. We saw how to register these virtual environments in Jupyter notebooks. We walked through two notebooks to develop baseline ML models, a simple NN, and a more advanced CNN for image classification. We looked at how to evaluate and deploy the model and use a simple REST service to host the model and respond to prediction requests. We tested the service with a sample Python client and some random images.

This service will be our main target when we describe adversarial attacks and defenses in the following chapters.

In the next chapter, we will discuss how traditional security applies to our new service and stage our first adversarial attack to demonstrate why traditional security is not enough to stop adversarial AI attacks.