arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

International Conference on Learning Representations · 会议 · Machine Learning

2026-08-26 至 2026-08-26 共收录 2
2606.24192 2026-08-26 cs.CV cs.AI cs.CL 版本更新

Co-occurring Associated REtained concepts in Diffusion Unlearning

扩散模型遗忘中的共现关联保留概念

Miso Kim, Georu Lee, Yunji Kim, Hoki Kim, Jinseong Park, Woojin Lee

机构 * Dongguk University-Seoul(东国大学-首尔) Chung-Ang University(Chung-Ang 大学) Korea Institute for Advanced Study(韩国高级研究院)

AI总结 针对扩散模型遗忘中误删良性共现概念的问题,提出CARE评分量化保留程度,并设计ReCARE框架通过自动构建CARE集实现稳健遗忘,平衡概念擦除、整体效用与共现概念保留。

Comments Accepted as a poster at ICLR 2026. Code available at this https URL (https://github.com/damilab/CARE)

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.13394 2026-08-26 cs.CV 版本更新

Spatial-DISE: A Unified Benchmark for Evaluating Spatial Reasoning in Vision-Language Models

空间-DisE:评估视觉语言模型空间推理能力的统一基准

Xinmiao Huang, Qisong He, Zhenglin Huang, Boxuan Wang, Zhuoyun Li, Guangliang Cheng, Yi Dong, Xiaowei Huang

机构 * School of Computer Science & informatics, University of Liverpool(计算机科学与信息学系,利物浦大学)

AI总结 本文提出Spatial-DISE基准,用于评估视觉语言模型的空间推理能力,通过四个象限分类和大规模数据集,揭示当前模型在多步骤多视角推理上的不足。

Comments ICLR 2026 Accepted Project Page: this https URL (https://shinmohuang.github.io/spatialdise_page/)

详情

展开后加载摘要…

URL PDF HTML 收藏