学习何时观察：一种解耦的课程学习框架，用于多模态推理中的战略感知
Learning When to Look: A Disentangled Curriculum for Strategic Perception in Multimodal Reasoning
AI总结：
本文提出了一种解耦课程学习框架，通过分阶段训练提升多模态推理中的抽象推理与战略视觉感知能力。
摘要：
多模态大语言模型（MLLMs）展现出显著潜力，但在复杂的、长链视觉推理任务中仍然脆弱。一个关键的失败模式是“视觉遗忘”，即模型在推理延伸时逐步失去视觉基础，这种现象恰当地描述为“想得更久，看得更少”。我们提出这种失败源于当前训练范式过早地将两种不同的认知技能纠缠在一起：（1）抽象逻辑推理（“如何思考”）和（2）战略视觉感知（“何时观察”）。这导致了基础的冷启动缺陷——削弱抽象推理——以及战略感知的不足，因为模型缺乏何时感知的策略。在本文中，我们提出了一种新的基于课程学习的框架，以解耦这些技能。首先，我们引入了一种解耦的监督微调（SFT）课程，先在纯文本数据上构建一个稳健的抽象推理基础，然后再通过一种新的感知基础的思维链（PG-CoT）范式将其锚定到视觉上。其次，我们通过将时间作为强化学习问题来解决战略感知的不足。我们设计了一种关键感知奖励，通过将感知动作与语言中的认知不确定性标记（例如，“等待”、“验证”）耦合，从而教模型何时观察，从而学习到一个自主的定位策略。我们的贡献包括对这两种缺陷的正式化以及开发了一种原理上、分两阶段的框架来解决这些问题，将模型从一种启发式驱动的观察者转变为一种战略、扎根的推理者。
英文摘要：
Multimodal Large Language Models (MLLMs) demonstrate significant potential but remain brittle in complex, long-chain visual reasoning tasks. A critical failure mode is "visual forgetting", where models progressively lose visual grounding as reasoning extends, a phenomenon aptly described as "think longer, see less". We posit this failure stems from current training paradigms prematurely entangling two distinct cognitive skills: (1) abstract logical reasoning "how-to-think") and (2) strategic visual perception ("when-to-look"). This creates a foundational cold-start deficiency -- weakening abstract reasoning -- and a strategic perception deficit, as models lack a policy for when to perceive. In this paper, we propose a novel curriculum-based framework to disentangle these skills. First, we introduce a disentangled Supervised Fine-Tuning (SFT) curriculum that builds a robust abstract reasoning backbone on text-only data before anchoring it to vision with a novel Perception-Grounded Chain-of-Thought (PG-CoT) paradigm. Second, we resolve the strategic perception deficit by formulating timing as a reinforcement learning problem. We design a Pivotal Perception Reward that teaches the model when to look by coupling perceptual actions to linguistic markers of cognitive uncertainty (e.g., "wait", "verify"), thereby learning an autonomous grounding policy. Our contributions include the formalization of these two deficiencies and the development of a principled, two-stage framework to address them, transforming the model from a heuristic-driven observer to a strategic, grounded reasoner. \textbf{Code}: \url{https://github.com/gaozilve-max/learning-when-to-look}.
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。
