Centroid Distance Distillation for Effective Rehearsal in Continual Learning
This work addresses catastrophic forgetting in continual learning, an incremental improvement for AI systems that need to learn sequentially without forgetting old tasks.
The paper tackles the problem of continual domain drift in rehearsal-based continual learning by introducing centroid distance distillation, which reduces sample bias and drift, achieving superior results on four datasets.
Rehearsal, retraining on a stored small data subset of old tasks, has been proven effective in solving catastrophic forgetting in continual learning. However, due to the sampled data may have a large bias towards the original dataset, retraining them is susceptible to driving continual domain drift of old tasks in feature space, resulting in forgetting. In this paper, we focus on tackling the continual domain drift problem with centroid distance distillation. First, we propose a centroid caching mechanism for sampling data points based on constructed centroids to reduce the sample bias in rehearsal. Then, we present a centroid distance distillation that only stores the centroid distance to reduce the continual domain drift. The experiments on four continual learning datasets show the superiority of the proposed method, and the continual domain drift can be reduced.