Improving Representation Learning Through Variational Autoencoding