鱼与熊掌不可兼得：概念纠缠限制了扩散模型的去学习能力
You Can't Have It Both Ways: Concept Entanglement Limits Diffusion Model Unlearning
浏览论文内容
中文总结 AI 辅助
本研究证明扩散模型概念去学习中目标与相关概念的重叠导致擦除与保留不可兼得，提出帕累托前沿评估框架，揭示现有十三种方法均无法同时满足强擦除与强保留。
中文摘要 AI 辅助
文本到图像扩散模型中的概念去学习旨在抑制目标概念（如马）的同时保留相关但不同的内容（如驴），然而现有方法要么在间接提示下泄露目标概念，要么明显损害其他概念。我们证明这些失败模式源于概念表征的几何结构，而非任何特定算法。通过将概念形式化为激活空间中的区域，我们证明了目标概念与其他概念之间的重叠程度为任何稳健擦除必须对其造成的损害设定了下界，且该权衡随重叠程度线性增长。在包括旨在保留非目标概念的方法在内的十三种去学习方法中，没有一种方法能同时实现强擦除和强邻居保留：STEREO几乎消除了间接泄露，但将邻居生成削减了超过75%，而稀疏推理时方法保留了邻居但存在泄露。损害随我们的重叠度量增加而增加，对STEREO而言是单调的；κ缩放关系在SDXL上复现，邻居选择性损害在FLUX上再次出现。对于纠缠概念，完美去学习是错误的追求目标；方法应在我们的定理建立的帕累托前沿上进行评估。
英文摘要
Concept unlearning in text-to-image diffusion models aims to suppress a target concept (e.g., \texttt{horse}) while preserving related but distinct content (e.g., \texttt{donkey}), yet existing methods either leak under indirect prompts or visibly degrade other concepts. We show that these failure modes stem from the geometry of concept representations rather than from any particular algorithm. Formalizing concepts as activation-space regions, we prove that the overlap between a target and other concepts lower-bounds the damage any robust erasure must inflict on them, with the trade-off scaling linearly in the degree of overlap. Across thirteen unlearning methods, including methods designed to preserve non-target concepts, no method achieves both strong erasure and strong neighbor preservation: STEREO nearly eliminates indirect leakage but cuts neighbor generation by more than 75\%, while sparse inference-time methods preserve neighbors but leak. Damage increases with our overlap measure, monotonically so for STEREO; the $κ$-scaling reproduces on SDXL, and neighbor-selective damage recurs on FLUX. Perfect unlearning is the wrong target for entangled concepts; methods should be evaluated on the Pareto frontier our theorem establishes.
发表机构
- University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-尚佩恩分校）
机构由 AI 辅助整理，请以论文原文为准。
