arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2604.07518cs.CL
收藏

分解、观察与推理:用于视觉语言模型的强化潜在推理

Decompose, Look, and Reason: Reinforced Latent Reasoning for VLMs

Emory University(埃默里大学);University of Michigan, Ann Arbor(密歇根大学安娜堡分校)

Mengdan Zhu, Senhao Cheng, Liang Zhao

AI总结:

本文提出DLR框架,通过动态分解查询、提取连续视觉潜在表示和基于 grounded rationales 推理,提升视觉语言模型在复杂视觉推理中的表现,实验表明其优于现有基线方法。

摘要:

视觉语言模型在复杂视觉推理中常因文本推理中的视觉信息丢失而遇到困难。现有方法或增加工具调用成本,或依赖局部补丁嵌入,难以提取多步骤推理中的语义。我们提出

英文摘要:

Vision-Language Models often struggle with complex visual reasoning due to the visual information loss in textual CoT. Existing methods either add the cost of tool calls or rely on localized patch-based embeddings that are insufficient to extract semantics in multi-step reasoning. We propose "Decompose, Look, and Reason" (DLR), a reinforced latent reasoning framework that dynamically decomposes queries into textual premises, extracts premise-conditioned continuous visual latents, and deduces answers through grounded rationales. We introduce a three-stage training pipeline and propose a novel Spherical Gaussian Latent Policy, to enable effective exploration in the latent space. Extensive experiments on vision-centric benchmarks show that DLR consistently outperforms strong baselines, including text-only, interleaved multimodal CoT, and latent reasoning methods, while providing superior stepwise interpretability.

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。

↑