In this chapter, we explained the main concept of CNNs and how to use them in Keras.
We saw why a convolutional layer is an efficient approach for problems where the inputs have high spatial correlation. We also saw the mathematics behind the convolutional layer, and how our filters are able to capture the features.
We discussed the need for pooling layers, softmax activation, and zero padding to avoid the shrinking of our images, especially for deep neural networks (DNNs).
We also saw how it's possible to debug our network to detect problems, checking the activation maps, filters, and saliency maps.
We discussed the various possible uses of CNN-like image classification and image detection, and how they are actually very flexible and can be used to solve many different tasks.
In the next chapter, we will focus on deep learning for natural language processing, but...