多思考，少看见？评估多模态推理模型中增强的幻觉现象
More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Models
AI总结：
本研究提出RH-AUC指标和RH-Bench基准，系统评估多模态推理模型在长推理链下的幻觉加剧现象，发现模型规模与训练数据类型影响推理与感知的平衡。
摘要：
测试时计算能力的提升使得多模态大语言模型能够生成更长的推理链，在多模态数学推理等任务上展现出强劲性能。然而，这种推理能力的提升往往伴随着幻觉现象的加剧：随着生成内容变长，模型倾向于偏离基于图像的内容，更多依赖语言先验。注意力分析表明，较长的推理链导致模型对视觉输入的关注度降低，这进而引发了幻觉。为系统研究这一现象，我们引入了RH-AUC指标，该指标量化了模型感知准确度随推理长度的变化，使我们能够评估模型在推理过程中是否保持视觉基础。我们还发布了RH-Bench，一个涵盖多种多模态任务的诊断基准，旨在评估推理能力与幻觉之间的权衡。我们的分析揭示：（i）较大的模型通常能在推理与感知之间取得更好的平衡；（ii）这种平衡受训练数据的类型和领域影响更大，而非其总体规模。这些发现强调了在评估框架中同时考虑推理质量与感知保真度的重要性。
英文摘要：
Test-time compute has empowered multimodal large language models to generate extended reasoning chains, yielding strong performance on tasks such as multimodal math reasoning. However, this improved reasoning ability often comes with increased hallucination: as generations become longer, models tend to drift away from image-grounded content and rely more heavily on language priors. Attention analysis shows that longer reasoning chains lead to reduced focus on visual inputs, which contributes to hallucination. To systematically study this phenomenon, we introduce RH-AUC, a metric that quantifies how a model's perception accuracy changes with reasoning length, allowing us to evaluate whether the model preserves visual grounding during reasoning. We also release RH-Bench, a diagnostic benchmark that spans a variety of multimodal tasks, designed to assess the trade-off between reasoning ability and hallucination. Our analysis reveals that (i) larger models typically achieve a better balance between reasoning and perception, and (ii) this balance is influenced more by the types and domains of training data than by its overall volume. These findings underscore the importance of evaluation frameworks that jointly consider both reasoning quality and perceptual fidelity.
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。
