分层近端重放：一种用于在线持续学习的近端点方法
Layerwise Proximal Replay: A Proximal Point Method for Online Continual Learning
AI总结：
针对在线持续学习中经验重放导致优化轨迹不稳定、影响准确率的问题，提出分层近端重放（LPR），通过修改优化几何、限制过去数据隐藏激活的渐进变化，在多种设置下稳定提升重放方法性能。
摘要：
在在线持续学习中，神经网络从不独立同分布的数据流中增量学习。几乎所有在线持续学习方法都采用经验重放，以同时防止灾难性遗忘和对过去数据的欠拟合。我们的工作展示了这种方法的一个局限性：使用经验重放训练的神经网络往往具有不稳定的优化轨迹，从而阻碍了其整体准确性。令人惊讶的是，即使重放缓冲区存储了所有先前的训练样本，这些不稳定性仍然存在，这表明该问题与灾难性遗忘是正交的。我们通过对优化几何进行简单修改来最小化这些不稳定性。我们的解决方案——分层近端重放（LPR）——在平衡来自新数据和重放数据的学习的同时，仅允许过去数据的隐藏激活发生渐进变化。我们证明，无论可用的重放内存量如何，LPR都能在多种问题设置中持续改进基于重放的在线持续学习方法。
英文摘要：
In online continual learning, a neural network incrementally learns from a non-i.i.d. data stream. Nearly all online continual learning methods employ experience replay to simultaneously prevent catastrophic forgetting and underfitting on past data. Our work demonstrates a limitation of this approach: neural networks trained with experience replay tend to have unstable optimization trajectories, impeding their overall accuracy. Surprisingly, these instabilities persist even when the replay buffer stores all previous training examples, suggesting that this issue is orthogonal to catastrophic forgetting. We minimize these instabilities through a simple modification of the optimization geometry. Our solution, Layerwise Proximal Replay (LPR), balances learning from new and replay data while only allowing for gradual changes in the hidden activation of past data. We demonstrate that LPR consistently improves replay-based online continual learning methods across multiple problem settings, regardless of the amount of available replay memory.
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。
