聚焦推理：面向视觉语言模型的有状态、基于动作的视觉聚焦
Foveated Reasoning: Stateful, Action-based Visual Focusing for Vision-Language Models
AI总结：
本文提出Foveated Reasoner，通过整合聚焦与推理机制，提升视觉语言模型在高分辨率图像下的效率与准确性，实验证明其在多种基准测试中表现优异。
摘要：
视觉语言模型受益于高分辨率图像，但增加的视觉token数量会带来高计算开销。人类通过聚焦解决这一矛盾：粗略的视角指导“看向何处”，而选择性获取的高清晰度证据细化“思考什么”。我们引入Foveated Reasoner，一种自回归的视觉语言框架，将聚焦与推理统一于单个解码轨迹中。从低分辨率视角开始，模型仅在需要时触发聚焦，从选定区域检索高分辨率证据，并将其注入相同的解码轨迹。我们通过两阶段流程训练该方法：冷启动监督以引导聚焦行为，随后通过强化学习共同改进证据获取和任务准确性，同时抑制简单的“看到一切”解决方案。实验表明，该方法学习了有效的聚焦策略，并在紧缩的视觉token预算下，于多个视觉语言基准测试中实现了更强的准确性。
英文摘要：
Vision-language models benefit from high-resolution images, but the increase in visual-token count incurs high compute overhead. Humans resolve this tension via foveation: a coarse view guides "where to look", while selectively acquired high-acuity evidence refines "what to think". We introduce Foveated Reasoner, an autoregressive vision-language framework that unifies foveation and reasoning within a single decoding trajectory. Starting from a low-resolution view, the model triggers foveation only when needed, retrieves high-resolution evidence from selected regions, and injects it back into the same decoding trajectory. We train the method with a two-stage pipeline: coldstart supervision to bootstrap foveation behavior, followed by reinforcement learning to jointly improve evidence acquisition and task accuracy while discouraging trivial "see-everything" solutions. Experiments show that the method learns effective foveation policies and achieves stronger accuracy under tight visual-token budgets across multiple vision-language benchmarks.
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。
