In previous chapters, we've learned how to use regular feed-forward network architectures to build neural networks. In a feed-forward neural network, we assume that there are interactions between the different input features. But we don't make any assumptions about the nature of these interactions. This is, however, not always the right thing to do.
When you work with complex data such as images, a feed-forward neural network won't do a very good job. This comes from the fact that we assume that there's an interaction between the inputs of our network. But we don't account for the fact that they are organized in a spatial way. When you look at the pixels in an image, there's a horizontal and vertical relationship between them. There's also a relationship between the colors in an image and the position...