Throughout this chapter, we're going to make use of a popular dataset called the MNIST database. This has been made available by Yann LeCun, Corinna Cortes, and Christopher Burges at http://yann.lecun.com/exdb/mnist.

The database gets its name from the fact that it was made by mixing two databases that contain black and white images of handwritten digits. It is an example of an ideal dataset that has been preprocessed and formatted nicely for us so that we can immediately start using it. When you download it, it is already divided into training and testing (validation) sets, with 60,000 labeled examples in the training set and 10,000 labeled examples in the test set.

Each image is exactly 28 x 28 pixels and contains a value from 1 to 255 (reflecting the pixel intensity or grayscale value). This greatly simplifies things for us, as it means that we can immediately put the image into a matrix/tensor and start training our models on it.