[论文] GeoReform: Reflective Formalization Evolution for Multimodal Geometry ...

论文概要

研究领域: ML
作者: Jialu Wang, Ruichen Zhang, Xiaoou Liu, Hua Wei, Tianlong Chen
发布时间: 2026-10-08
arXiv: 2610.12391

中文摘要

多模态大语言模型（MLLM）常常难以识别和利用图中的几何关系。近期方法通过将几何实体、关系和约束转化为显式文本表示来解决这一挑战，但有效的形式化绝非易事：在 Geometry3K 上，结构注入修复了 28 个错误却在 200 个样本中引入了 13 个新错误。冗余的关系会分散模型注意力，而对图中元素的模糊引用会导致约束被错误应用。这说明关键挑战不在于提取更多几何事实，而在于将它们组织成支持下游推理的表示。为此我们提出 GeoReform——一个反思式形式化演化框架，将形式化视为可优化的策略而非固定的解析器输出。GeoReform 执行完整推理流水线，收集失败的 rollout，诊断当前表示的缺陷，并变异策略以更好地选择、定位、分组和呈现几何实体、关系、约束和目标。在 Geometry3K 上，GeoReform 将 Qwen3VL-2B 的准确率从 42.0% 提升到 56.0%。大量实验表明，有效的形式化对改进多模态几何推理至关重要。

原文摘要

Multimodal large language models (MLLMs) often struggle to identify and use geometric relations in diagrams. Recent methods address this challenge by converting geometric entities, relations, and constraints into explicit textual representations for the model to reason over. However, effective formalization is highly non-trivial: on Geometry3K, structure injection fixes 28 errors but introduces 13 new ones among 200 examples. Redundant relations can distract the model, while ambiguous references to diagram elements can lead it to apply constraints incorrectly. This suggests that the key challenge is not merely extracting more geometric facts, but organizing them into representations that support downstream reasoning. To fully exploit the power of formalization, we further propose GeoReform...

自动采集于 2026-10-12

#论文 #arXiv #ML #小凯