可视化思维：概念图使大型多模态模型具备稳健规划能力
Visualizing Thought: Conceptual Diagrams Enable Robust Planning in LMMs
AI总结：
本文提出 Visual Thinking 框架，让 LMM 借助自生成概念图、Graph-of-Thought、束搜索与回溯进行多链推理，在 PDDL 规划任务上显著优于纯文本方法及 o1-preview。
摘要：
人类推理依赖于构建和操作心理模型，即用于理解和解决问题的情境的简化内部表征。概念图（例如为辅助推理而绘制的草图）将这些心理模型外化，抽象掉无关细节，以高效捕捉实体之间的交互方式。相比之下，大型语言模型（LLM）和大型多模态模型（LMM）主要通过文本进行推理，这限制了它们在复杂多步骤任务上的有效性。本文提出 Visual Thinking，这是一个可泛化的框架，使 LMM 能够通过多条自生成概念图链进行推理，从而显著增强其组合规划能力。除任务的自然语言描述外，该方法不需要任何人工输入。它在优化的 Graph-of-Thought 推理框架内整合文本推理与图示推理，并通过束搜索和按深度回溯加以增强。在多个具有挑战性的 PDDL 规划领域上评估时，该方法大幅提升了 LMM 的性能（例如 GPT-4o 在 Blocksworld 中从 35.5% 提升至 90.2%），并始终优于仅基于文本的搜索式推理方法。在解深度最高达 40 的更困难领域中，它还超过了 o1-preview 推理模型（例如在 Floor Tiles 中提升 16 个百分点）。这些结果证明了概念图作为 LMM 中推理媒介的强大能力。
英文摘要：
Human reasoning relies on constructing and manipulating mental models -- simplified internal representations of situations used to understand and solve problems. Conceptual diagrams (e.g., a sketch drawn to aid reasoning) externalize these mental models, abstracting irrelevant details to efficiently capture how entities interact. In contrast, Large Language Models (LLMs) and Large MultiModal Models (LMMs) predominantly reason through text, limiting their effectiveness on complex multi-step tasks. In this paper, we propose Visual Thinking, a generalizable framework that enables LMMs to reason through multiple chains of self-generated conceptual diagrams, significantly enhancing their combinatorial planning capabilities. Our approach requires no human input beyond the natural language description of the task. It integrates textual and diagrammatic reasoning within an optimized Graph-of-Thought inference framework, enhanced by beam search and depth-wise backtracking. Evaluated on multiple challenging PDDL planning domains, our method substantially improves LMM performance (e.g., GPT-4o: 35.5% -> 90.2% in Blocksworld) and consistently outperforms text-only search-based inference methods. On more difficult domains with solution depths up to 40, it also surpasses the o1-preview reasoning model (e.g., 16 percentage points improvement in Floor Tiles). These results demonstrate the power of conceptual diagrams as a reasoning medium in LMMs.
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。
