arXiv preprint / cs.CV

The Evidence Anchoring Boundary: Where Visual Interventions Matter and When They Help in Multimodal LLMs

作者：Qiji Zhou, Junyi Hu, Peiyu Wang, Yue Zhang · 单位：Westlake University, Duke University, Institute of Automation, Chinese Academy of Sciences · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-12

2026-10-12视觉语言学习理论
新标签打开 AlphaXiv
The Evidence Anchoring Boundary: Where Visual Interventions Matter and When They Help in Multimodal LLMs figure
AlphaXiv 中文概览（可滚动查看）