arXiv：cs.AI· Yang Xiao, Kaiyuan Deng, Gen Li, Zinan Ling, Zhiwen Fan, Xiaolong Ma, Bo Hui·· 17 小时前
DMU：通过域流形遗忘解耦扩散模型中的对象-风格组合概念擦除
Compositional Concepts Erasing: Disentangling Object-Style via Domain Manifold Unlearning
AI 导读
针对扩散模型遗忘学习中对象-风格组合概念难以擦除且多次请求易致灾难性遗忘的问题，研究者提出 Domain Manifold Unlearning（DMU），用从扩散中间特征提取的 EMA 更新特征代理表示保留目标对象方向，实现对象内特定域组件的选择性遗忘。实验显示 DMU 显著缓解灾难性遗忘，在批量遗忘与顺序遗忘上均优于现有 SOTA 方法。
正文
Abstract:Diffusion models pose significant risks regarding the generation of illicit content and copyright violations. This has catalyzed the development of machine unlearning to erase harmful concepts or copyrighted styles from a model's generative repertoire without compromising its general performance. However, previous unlearning approaches on diffusion models have largely focused on either object-level or style-level removal and failed to address the challenge of object-style combination unlearning. At the same time, multiple unlearning requests often lead to catastrophic forgetting. To address these challenges, we propose Domain Manifold Unlearning (DMU), which uses an EMA-updated feature proxy extracted from intermediate diffusion features to represent the retained target-object direction. As a lightweight estimating representation of an object, DMU enables selective unlearning of domain-specific components in a specific object while preserving other semantic representations, thereby significantly enhancing robustness in multiple combination unlearning scenarios. Extensive experiments show that DMU significantly mitigates catastrophic forgetting, outperforming state-of-the-art methods of batch unlearning and sequential unlearning.
| Comments: | 14 pages |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.14514 [cs.CV] |
| (or arXiv:2610.14514v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.14514 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yang Xiao [view email]
[v1]
Sat, 10 Oct 2026 03:22:31 UTC (23,478 KB)
来源：arXiv：cs.AI · arxiv.org