Isn't that the same material covered on Andrew Ng's Coursera course "Machine Learning", down to the training data?
The training data is the MNIST dataset released by NIST some time ago, anyone is allowed to use it. It truly is no surprise to see it here, as it is very a very commonly used dataset in ML tutorials/books. It receives some discussion in Artificial Intelligence: A Modern Approach by Russel and Norvig, and even in the Theano getting started tutorials.
In fact, if the goal of the book is to educate people on the field then I would say it's definitely best to use the standard benchmarks. It lets readers relate what's in the book to the literature, should they desire, and just like in academia, it lends credence to the author's statements. I've seen people take a lot of flak for publishing writings that don't use the standard datasets, but make claims of progress.