发表机构
The University of Queensland; Monash University（昆士兰大学; 莫纳什大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对现有域遗忘方法仅对已见类别有效的问题，提出开放词汇域遗忘协议及外科手术式参数编辑框架，通过Fisher信息掩码与目标流形散射实现类别无关的域擦除，在多个基准上以少量样本大幅提升开放词汇泛化性能。
AI 中文摘要
视觉-语言模型（VLMs）展现出卓越的零样本泛化能力，然而它们常常编码了不需要或危险的风格化域，例如医学AI中理想化的教科书图表或自动驾驶中的卡通车辆。近似域遗忘（ADU）旨在选择性地擦除模型对目标视觉域的识别能力，同时保持对剩余域的准确性。然而，现有的ADU方法在一种有缺陷的封闭词汇假设下运作：它们仅在遗忘微调阶段所见到的特定对象类别上评估遗忘效果。因此，这些方法并未真正遗忘该域本身；它们只是过拟合到已见的类别-域对，使得该域对于未见类别仍然易于识别，从而提供了一种虚假的移除感。我们认为真正的域擦除必须是类别无关的。为解决此问题，我们形式化了开放词汇域遗忘（OVDU），这是一个严格的协议，要求域遗忘必须迁移到保留类别上。为应对OVDU挑战，我们提出了一种外科手术式的参数编辑框架。首先，一个Fisher信息掩码隔离出域敏感权重，从数学上保护基础的零样本泛化能力。其次，我们的目标流形散射（TMS）目标利用基于偏好的挖掘来局部散射遗忘域的风格几何结构。在PACS、OfficeHome和DomainNet上的评估表明，我们的方法在开放词汇泛化方面大幅超越了现有基线。关键的是，它展现了卓越的样本效率，仅用4次样本就超越了8次样本基线的峰值结果。
英文摘要
Vision-Language Models (VLMs) exhibit remarkable zero-shot generalization, yet they often encode unwanted or hazardous stylistic domains such as idealized textbook diagrams in medical AI or cartoon vehicles in autonomous driving. Approximate Domain Unlearning (ADU) aims to selectively erase a model's recognition of a target visual domain while preserving accuracy on the remaining domains. However, existing ADU methods operate under a flawed closed-vocabulary assumption: they evaluate unlearning solely on the specific object classes seen during the unlearning fine-tuning phase. Consequently, these methods do not unlearn the domain itself; they merely overfit to seen class-domain pairs, leaving the domain easily recognizable for unseen classes and providing a false sense of removal. We argue that true domain erasure must be class-agnostic. To address this, we formalize Open-Vocabulary Domain Unlearning (OVDU), a rigorous protocol that mandates domain forgetting must transfer to held-out classes. To solve the OVDU challenge, we propose a surgical parameter-editing framework. First, a Fisher Information mask isolates domain-sensitive weights, mathematically protecting foundational zero-shot generalization. Second, our Targeted Manifold Scattering (TMS) objective uses preference-based mining to locally scatter the forget domain's stylistic geometry. Evaluated across PACS, OfficeHome, and DomainNet, our method vastly improves open-vocabulary generalization over existing baselines. Crucially, it delivers exceptional sample efficiency, outperforming peak 8-shot baseline results with only 4 shots.
CommentsAccepted in NeurIPS 2026
