Gradient Based Memory Editing for Task-Free Continual Learning
Xisen Jin, Junyi Du, Xiang Ren · arXiv (Cornell University) · 2021
Prior work on continual learning often operate in a “task-aware” manner, by assuming that the task boundaries and identifies of the data examples are known at all times. While in practice, it is rarely the case that such information are exposed to the methods (i.e., thus called “task-free”)–a setting that is relatively underexplored. Recent attempts on task-free continual learning build on previous memory replay methods and focus on developing memory construction and replay strategies such that model performance over previously seen examples can be best retained. In this paper, looking from a complementary angle, we propose a novel approach to “edit” memory examples so that the edited memory can better retain past performance when they are replayed. We use gradient updates to edit memory examples so that they are more likely to be “forgotten” in the future. Experiments on five benchmark datasets show the proposed method can be seamlessly combined with baselines to significantly improve the performance.