立桩定界：结构忠实的实例遗忘
Stake the Points: Structure-Faithful Instance Unlearning
把「语义锚点 + 结构感知对齐」作为可复用的遗忘正则手段：在删除指定样本时，用 CLIP 文本属性锚定保留集的知识结构，可直接迁移到人脸识别、检索模型的隐私删除与版权撤回场景。
原文依据 we propose a novel structure-faithful framework that introduces stakes, i.e., semantic anchors that serve as reference points to maintain the knowledge structure
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论机器遗忘（MU）旨在应对预训练模型中的隐私风险。其核心目标是消除指定数据的影响，同时保持所保留知识的效用。实现这一目标需要维持保留实例之间的语义关系，而现有研究往往忽视了这一点。我们观察到，若缺乏这种保持，模型会出现渐进式的结构崩塌，从而同时损害删除与保留之间的平衡。在本工作中，我们提出了一种新颖的结构忠实框架，引入「桩」（stakes），即作为参考点以维持知识结构的语义锚点。借助这些锚点，我们的框架能够刻画并稳定知识在语义层面的组织方式。具体而言，我们由语义编码器（如 CLIP）编码的语言驱动属性描述来实例化这些锚点。我们通过结构感知对齐与正则化来强制保持知识结构：前者围绕锚点，使遗忘前后保留知识的组织方式相互对齐；后者则约束对结构关键参数的更新。在图像分类、检索与人脸识别上的结果表明，性能平均提升 32.9%、22.5% 与 19.3%，在平衡删除—保留权衡的同时增强了泛化能力。
英文原文摘要 1341 字符
Machine unlearning (MU) addresses privacy risks in pretrained models. The main goal of MU is to remove the influence of designated data while preserving the utility of retained knowledge. Achieving this goal requires preserving semantic relations among retained instances, which existing studies often overlook. We observe that without such preservation, models suffer from progressive structural collapse, undermining both the deletion-retention balance. In this work, we propose a novel structure-faithful framework that introduces stakes, i.e., semantic anchors that serve as reference points to maintain the knowledge structure. By leveraging these anchors, our framework captures and stabilizes the semantic organization of knowledge. Specifically, we instantiate the anchors from language-driven attribute descriptions encoded by a semantic encoder (e.g., CLIP). We enforce preservation of the knowledge structure via structure-aware alignment and regularization: the former aligns the organization of retained knowledge before and after unlearning around anchors, while the latter regulates updates to structure-critical parameters. Results from image classification, retrieval, and face recognition show average gains of 32.9%, 22.5%, and 19.3% in performance, balancing the deletion-retention trade-off and enhancing generalization.