ClueTracer: 问题到视觉线索追踪用于无训练 hallucination 抑制在多模态推理
ClueTracer: Question-to-Vision Clue Tracing for Training-Free Hallucination Suppression in Multimodal Reasoning
AI总结：
ClueTracer通过问题到视觉线索追踪，无训练抑制多模态推理中的幻觉，提升推理和非推理任务性能。
摘要：
大型多模态推理模型通过显式的长链推理解决具有挑战性的视觉问题：它们从图像中收集视觉线索并解码为文本标记。然而，这种能力也增加了幻觉，即模型生成的内容没有得到输入图像或问题的支持。为了理解这种失败模式，我们识别出"推理漂移"：在线索收集过程中，模型过度关注与问题无关的实体，稀释了对任务相关线索的关注，并逐渐使推理轨迹与视觉基础脱节。因此，许多为非推理模型开发的推理时间定位或干预方法在推理设置中无法定位真正的线索。受这些见解启发，我们引入了 ClueRecall，一个评估视觉线索检索的指标，并提出了 ClueTracer，一种无训练、无参数、无架构依赖的插件，用于抑制幻觉。ClueTracer从问题出发，追踪关键线索如何沿着模型的推理路径（问题→输出→视觉标记）传播，从而定位任务相关区域并抑制对无关区域的虚假关注。值得注意的是，在没有任何额外训练的情况下，ClueTracer在推理基准上提高了所有推理架构（包括R1-OneVision、Ocean-R1、MM-Eureka等）的性能，提升幅度为1.21倍。当转移到非推理设置时，它产生了1.14倍的收益。
英文摘要：
Large multimodal reasoning models solve challenging visual problems via explicit long-chain inference: they gather visual clues from images and decode clues into textual tokens. Yet this capability also increases hallucinations, where the model generates content that is not supported by the input image or the question. To understand this failure mode, we identify \emph{reasoning drift}: during clue gathering, the model over-focuses on question-irrelevant entities, diluting focus on task-relevant cues and gradually decoupling the reasoning trace from visual grounding. As a consequence, many inference-time localization or intervention methods developed for non-reasoning models fail to pinpoint the true clues in reasoning settings. Motivated by these insights, we introduce ClueRecall, a metric for assessing visual clue retrieval, and present ClueTracer, a training-free, parameter-free, and architecture-agnostic plugin for hallucination suppression. ClueTracer starts from the question and traces how key clues propagate along the model's reasoning pathway (question $\rightarrow$ outputs $\rightarrow$ visual tokens), thereby localizing task-relevant patches while suppressing spurious attention to irrelevant regions. Remarkably, \textbf{without any additional training}, ClueTracer improves all \textbf{reasoning} architectures (including \texttt{R1-OneVision}, \texttt{Ocean-R1}, \texttt{MM-Eureka}, \emph{etc}.) by $\mathbf{1.21\times}$ on reasoning benchmarks
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。
