AI 中文总结
研究类增量学习中合成旧数据与真实新数据混合导致性能下降的问题，提出DREAM模型，利用无训练生成器合成旧数据，通过子空间校正、正交投影及真实锚定原型正则化消除域捷径并加强语义对齐，取得最优性能。
AI 中文摘要
类增量学习（CIL）要求模型持续获取新知识并避免灾难性遗忘。示例重放虽有效，但存在隐私和存储问题，生成重放应运而生，利用冻结的预训练文本到图像（T2I）模型合成旧数据且无需额外训练。然而，增量训练中直接混合合成旧类数据和真实新类数据会导致性能显著下降，这源于“域捷径”，模型依赖域判别特征而非语义类线索。为解决此问题，我们提出DREAM（域正则化无示例对齐模型），使用无训练生成器合成旧类数据，通过子空间校正和正交投影消除域捷径，同时通过真实锚定原型正则化加强语义对齐。在4个数据集上的大量实验表明，DREAM优于现有的无示例CIL方法并实现了当前最优性能。
英文摘要
Class-incremental learning (CIL) requires models to continuously acquire new knowledge while avoiding catastrophic forgetting. While exemplar replay is effective, it raises concerns regarding privacy and storage. Thus, generative replay has emerged as a viable alternative, synthesizing old data using frozen pretrained text-to-image (T2I) models without any extra training. However, we observe that directly mixing synthetic old-class data with real new-class data during incremental training leads to significant performance degradation. This issue stems from a "domain shortcut", where models rely on domain-discriminative features instead of semantic class cues. To address this, we propose DREAM ($\underline{\mathbf{D}}$omain-$\underline{\mathbf{R}}$egularized $\underline{\mathbf{E}}$xemplar-free $\underline{\mathbf{A}}$lignment $\underline{\mathbf{M}}$odel), which uses a training-free generator to synthesize old-class data and eliminates domain shortcut via subspace rectification and orthogonal projection, while reinforcing semantic alignment through real-anchored prototype regularization. Extensive experiments on 4 datasets demonstrate that DREAM outperforms existing exemplar-free CIL methods and achieves state-of-the-art performance. Our source code is available at https://github.com/Light-ZhangTao/DREAM.
CommentsAccepted by ICML 2026
