CNNs have been seminal in solving many computer vision tasks. In this chapter, we learned about how these networks differ from our basic feedforward networks, what their structures are, and how we can utilize them. CNNs are primarily used for computer vision tasks, although they can be adapted for use in other unstructured domains, such as natural language processing and audio signal processing.
CNNs are made up of convolutional layers interspersed with pooling layers, all of which output to a fully connected layer. CNNs iterate over images using filters. Filters have a size and a stride, which is how quickly they iterate over an input image. Input consistency can be better guaranteed by utilizing the zero padding technique.
In the next chapter, we'll learn about another important class of networks, called Recurrent Neural Networks.