跳到正文
arXiv cs.AI· Genglin Liu, Muye Zhang, Krishnamurthy Viswanathan, Nichole J. Hansen, Bla\v{z} Bratani\v{c}, Nathan L Clement, Shalini Ghosh, Ariel Fuxman·· 2 小时前AI 评分41

分歧引导的智能体数据筛选方法用于多模态安全评估

Disagreement-Guided Agentic Data Curation for Multimodal Safety

AI 导读

一项 arXiv 研究提出用标注分歧作为信号来筛选多模态安全边界样本的智能体数据筛选流程,生成政策条件化图像、按分歧选择样例并作 in-context 检索示例。

正文 · 原文

View PDF HTML (experimental)

Abstract:Written safety policies in multimodal settings define categories whose visual boundaries span many contexts. We study disagreement as a signal for identifying useful synthetic boundary cases. Our pipeline generates policy-conditioned images, selects cases with target--rater or inter-rater disagreement, conditionally escalates unresolved cases, and retrieves the resulting rater-labeled examples as in-context demonstrations. On a failure-enriched multimodal safety benchmark, policy grounding reduces false negative rate (FNR) from 0.875 to 0.413 and raises F1 from 0.220 to 0.731; retrieval from 50 disagreement-selected examples further reduces FNR to 0.245, compared with 0.297 from a 350-example Regular pool. The selected memory also improves a different target-model family, with a smaller gain on a proprietary benchmark with a refined policy. In a blinded human study, experts disagree on 24% of disagreement-guided images and 17% of baseline-curated images. These findings support traceable, disagreement-guided curation for multimodal safety evaluation and example selection.
Comments: 9 pages; published in 2026 IEEE International Conference on Big Data
Subjects: Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
Cite as: arXiv:2607.14256 [cs.AI]
  (or arXiv:2607.14256v2 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2607.14256

arXiv-issued DOI via DataCite

Submission history

From: Genglin Liu [view email]
[v1] Wed, 15 Jul 2026 18:13:05 UTC (9,713 KB)
[v2] Wed, 18 Nov 2026 04:04:22 UTC (2,578 KB)

来源:arXiv cs.AI · arxiv.org