Retrospective Adversarial Replay for Continual Learning

Neural Information Processing Systems 

To avoid these problems, this paper proposes a method, "Retrospective Adversarial