Drift and Dependence: Layer-wise Information-Theoretic Bounds for Replay-Based Continual Learning
漂移与依赖:基于重放的持续学习的分层信息论边界
机构 * School of Computer Science and Technology, Xi’an Jiaotong University(西安交通大学计算机科学与技术学院) ; Department of Computer Science and Technology, Tsinghua University(清华大学计算机科学与技术系)
AI总结 该研究提出分层信息论框架,分解重放式持续学习的泛化差距,通过Wasserstein松弛和SGLD实例化实现可操作的边界,经实验验证了相关预测。