Convolutional Neural Networks

This section will focus on the CNN's architecture. Though CNN is a topic that might not be covered fully in a single chapter, we will focus on the important elements that a CNN has for you to get started with CNNs easily. During our discussion, we will use the Keras package to create a CNN using the sample MNIST dataset that the package has.

The question that comes to mind when we hear the term CNN is, why CNN? We will try to answer this question with a short explanation.

Why CNN?

We discussed feed-forward neural networks in an earlier section. Though they are powerful, one of their main disadvantages is that an FNN ignores the structure of the input data. All data feed to the network has to be first converted into a 1D numerical array. However, for higher-dimensional arrays such as in an image, it gets difficult to deal with such conversion. It is essential to preserve the structure of images, as there is a lot of hidden information stored inside the, this is...