Disentangled Representations in Neural Models