校准置信度的幻象：视觉语言模型中言语化置信度的轨迹无关性
The Mirage of Calibrated Confidence: Trajectory-Independence of Verbalized Confidence in Vision-Language Models
浏览论文内容
中文总结 AI 辅助
本研究揭示视觉语言模型的言语化置信度与推理轨迹无关，提出轨迹接地分数（TGS）及TGS-Bench基准，以检测并弥补现有校准评估的盲点。
中文摘要 AI 辅助
一个校准良好的视觉语言模型（VLM）可以反复自我纠正，说“等等，我应该重新检查”，得出错误答案，却仍然报告高置信度。我们发现，这是因为在我们评估的视觉语言模型和校准方法中，言语化置信度在很大程度上是轨迹无关的。我们通过三个互补的视角来审视这一点：内容变化、令牌掩蔽以及模型自身的犹豫标记。我们表明，置信度对推理轨迹实际包含的内容不够敏感，而校准训练可能矛盾地加剧这种脱节。由于现有的指标如ECE和AUROC无法检测到这个问题，我们提出了轨迹接地分数（TGS），包含两种互补形式：TGS-self，它比较模型在有无自身轨迹访问情况下的置信度；以及TGS-pair，它测试模型是否在视觉、推理和答案轴上对正确轨迹赋予比有缺陷轨迹更高的置信度。我们提出了TGS-Bench，一个涵盖10个基准的模型无关套件，具有受控的好/坏轨迹对，并表明传统的校准排名与轨迹接地排名存在分歧，揭示了当前评估实践中的一个盲点。
英文摘要
A calibrated Vision-Language Model (VLM) can repeatedly self-correct, say "Wait, I should recheck," arrive at the wrong answer, and still report high confidence. We find that this occurs because verbalized confidence is largely trajectory-independent in the VLMs and calibration methods we evaluate. We examine this through three complementary lenses: content variation, token masking, and the model's own hesitation markers. We show that confidence is insufficiently sensitive to what the reasoning trajectory actually contains, and that calibration training can paradoxically worsen this disconnect. Since existing metrics like ECE and AUROC cannot detect this problem, we propose the Trajectory-Grounding Score (TGS) in two complementary forms: TGS-self, which compares confidence with and without access to the model's own trajectory, and TGS-pair, which tests whether the model assigns higher confidence to correct trajectories than to flawed ones along the vision, reasoning, and answer axes. We propose TGS-Bench, a model-agnostic suite spanning 10 benchmarks with controlled good/bad trajectory pairs, and show that conventional calibration rankings diverge from trajectory-grounding rankings, exposing a blind spot in current evaluation practice.
发表机构
- Chung-Ang University（中央大学）
- Van Lang University（文朗大学）
机构由 AI 辅助整理，请以论文原文为准。
