Stacked autoencoder exercise from http://ufldl.stanford.edu/ - View it on GitHub