PulseAugur
实时 12:39:32
English(EN) Look Again Before You Abstain:Budgeted Conformal Evidence Acquisition for Reliable Vision-Language Model

新方法通过证据获取解决视觉语言模型幻觉问题

研究人员开发了一种名为带预算共形证据获取(BCEA)的新方法,以解决大型视觉语言模型(LVLMs)中的幻觉问题。传统的需要弃权预测以维持准确性的方法效率极低,常常弃权超过80%的声明。BCEA提供了一种更细致的方法,允许模型在计算预算内进行回答、弃权或获取额外的视觉证据,从而恢复统计保证并提高覆盖率。 AI

影响 这项研究通过智能地获取更多数据而不是简单地弃权预测,为确保视觉语言模型的准确性提供了一种更有效的方法。

排序理由 该集群包含一篇学术论文,详细介绍了一种提高视觉语言模型可靠性的新方法。

在 arXiv cs.CV 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →

报道来源 [2]

  1. arXiv cs.CV TIER_1 English(EN) · Jian Xu, Delu Zeng, John Paisley, Qibin Zhao ·

    Look Again Before You Abstain:Budgeted Conformal Evidence Acquisition for Reliable Vision-Language Model

    arXiv:2606.16667v1 Announce Type: new Abstract: Large vision-language models (LVLMs) hallucinate: they assert visual details that the image does not support. A principled remedy is selective prediction with a distribution-free guarantee-verify each claim and abstain when the clai…

  2. arXiv cs.CV TIER_1 English(EN) · Qibin Zhao ·

    Look Again Before You Abstain:Budgeted Conformal Evidence Acquisition for Reliable Vision-Language Model

    Large vision-language models (LVLMs) hallucinate: they assert visual details that the image does not support. A principled remedy is selective prediction with a distribution-free guarantee-verify each claim and abstain when the claim is not grounded, so that the hallucination rat…