In this chapter, we will modify our one-dimensional neural network by adding convolutions to produce our first actual convolutional (2D) neural network and use it to categorize black and white (e.g., MNIST) images again.
Convolutions
Convolutions are a deep area of computer vision theory. At a high level we might think of taking an input image and producing another output image:
[cat] --> [magic black box] --> [dog]
Broadly, for any input image there’s a way to convert it to the target image. At the simplest level we might destroy the source image (e.g., multiply ...
