Developing our target AI service with CNNs

MLPs are relatively simple. The main NN in our playground will be CNN, a type of deep NN (DNN) popular in image and object recognition tasks. We will use CIFAR-10, a public dataset bundled with Keras, to create a sample CNN, save it, and then deploy it for predictions.

CIFAR-10 contains 60,000 32 x 32 images for 10 classes. For more information, see https://keras.io/api/datasets/cifar10/.

We have included a detailed explanation of the steps, acting as a guide to how each step works. We will describe the steps here, but we advise you to walk through the Jupyter notebook and read the comments to understand better how each step works. You can find the notebook in our GitHub repository: https://github.com/PacktPublishing/Adversarial-AI---Attacks-Mitigations-and-Defense-Strategies/blob/main/ch2/simple-cnn-cifar10.ipynb.

We have also added a deployment function and code to run predictions (inference) as a REST service and test it with some...