Representation Learning With Adversarial Latent Autoencoders