Adversarial Continual Learning