VL-Uncertainty：通过不确定性估计检测大型视觉-语言模型中的幻觉
VL-Uncertainty: Detecting Hallucination in Large Vision-Language Model via Uncertainty Estimation
AI总结：
针对 LVLM 幻觉检测依赖标注且成本高的问题，提出 VL-Uncertainty，通过扰动视觉/文本提示、语义聚类响应并计算分布熵来估计不确定性，在四个基准和 10 个模型上显著优于基线。
摘要：
鉴于大型视觉-语言模型（LVLM）相比单模态 LLM 处理更高的信息负载，检测 LVLM 幻觉需要更多人力与时间开销，并因此引发更广泛的安全担忧。本文提出 VL-Uncertainty，这是首个基于不确定性的 LVLM 幻觉检测框架。与大多数需要真值或伪标注的现有方法不同，VL-Uncertainty 将不确定性作为一种内在指标。我们通过分析语义等价但经过扰动的提示（包括视觉和文本数据）之间的预测方差来度量不确定性。当 LVLM 高度自信时，它们会对语义等价的查询给出一致响应；然而，当不确定时，目标 LVLM 的响应会变得更加随机。考虑到语义相似但措辞不同的答案，我们根据 LVLM 响应的语义内容对其进行聚类，随后计算聚类分布熵，作为检测幻觉的不确定性度量。我们在四个基准上对 10 个 LVLM 进行了大量实验，涵盖自由形式和多选任务，结果表明 VL-Uncertainty 在幻觉检测方面显著优于强基线方法。
英文摘要：
Given the higher information load processed by large vision-language models (LVLMs) compared to single-modal LLMs, detecting LVLM hallucinations requires more human and time expense, and thus rise a wider safety concerns. In this paper, we introduce VL-Uncertainty, the first uncertainty-based framework for detecting hallucinations in LVLMs. Different from most existing methods that require ground-truth or pseudo annotations, VL-Uncertainty utilizes uncertainty as an intrinsic metric. We measure uncertainty by analyzing the prediction variance across semantically equivalent but perturbed prompts, including visual and textual data. When LVLMs are highly confident, they provide consistent responses to semantically equivalent queries. However, when uncertain, the responses of the target LVLM become more random. Considering semantically similar answers with different wordings, we cluster LVLM responses based on their semantic content and then calculate the cluster distribution entropy as the uncertainty measure to detect hallucination. Our extensive experiments on 10 LVLMs across four benchmarks, covering both free-form and multi-choice tasks, show that VL-Uncertainty significantly outperforms strong baseline methods in hallucination detection.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
