Selective attention-driven modulation can effectively enhance the performance and robustness of continual learning models by leveraging the forgetting-free behavior of saliency prediction.
