GRASP：一种用于高效在线持续学习的 rehearsal 策略
GRASP: A Rehearsal Policy for Efficient Online Continual Learning
AI总结：
本文提出 GRASP rehearsal 策略，通过先选原型性强的简单样本再逐步选更难样本，以极低额外开销提升在线持续学习效率，在 ImageNet 和文本分类任务上优于多种现有策略。
摘要：
深度神经网络（DNN）中的持续学习（CL）涉及从不断增长的数据流中增量地积累知识。CL 面临的一个主要挑战是，非平稳数据流会导致对先前所学能力的灾难性遗忘。一种流行的解决方案是 rehearsal：将过去的观测存储在缓冲区中，然后从缓冲区采样以更新 DNN。以类别平衡方式进行均匀采样非常有效，而更好的样本选择策略一直难以实现。在此，我们提出一种新的样本选择策略，称为 GRASP，它首先选择最具原型性（简单）的样本，然后逐步选择原型性较弱（更难）的样本。与均匀选择相比，GRASP 几乎没有额外的计算或内存开销，使其能够扩展到大型数据集。与 17 种其他 rehearsal 策略相比，GRASP 在 ImageNet 上的 CL 实验中取得了更高的准确率。与均匀平衡采样相比，GRASP 以少 40% 的更新次数达到了相同的性能。我们还表明，GRASP 对五个文本分类数据集上的 CL 是有效的。
英文摘要：
Continual learning (CL) in deep neural networks (DNNs) involves incrementally accumulating knowledge in a DNN from a growing data stream. A major challenge in CL is that non-stationary data streams cause catastrophic forgetting of previously learned abilities. A popular solution is rehearsal: storing past observations in a buffer and then sampling the buffer to update the DNN. Uniform sampling in a class-balanced manner is highly effective, and better sample selection policies have been elusive. Here, we propose a new sample selection policy called GRASP that selects the most prototypical (easy) samples first and then gradually selects less prototypical (harder) examples. GRASP has little additional compute or memory overhead compared to uniform selection, enabling it to scale to large datasets. Compared to 17 other rehearsal policies, GRASP achieves higher accuracy in CL experiments on ImageNet. Compared to uniform balanced sampling, GRASP achieves the same performance with 40% fewer updates. We also show that GRASP is effective for CL on five text classification datasets.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
