Supervised autoencoders: Improving generalization performance with unsupervised regularizers