Evidence-RL：面向证据密集型视觉推理的反事实证据解缠方法
原标题：Evidence-RL: Towards Evidence-intensive Visual Reasoning
AI 摘要
Hugging Face Daily Papers 发布了一篇题为《Evidence-RL: Towards Evidence-intensive Visual Reasoning》的论文。论文提出了一种名为反事实证据解缠（CED）的训练时证据审计方法，用于增强视觉语言模型（VLM）的接地能力。CED 通过中和对象中心的证据区域并比较支持度下降，结合 GRPO 奖励依赖证据路径的正确答案，无需问题特定的证据标注且不增加推理开销。实验表明，CED 在九个公共基准和四个骨干网络上优于先前的基于强化学习的后训练方法。
正文节选
Evidence-RL: Towards Evidence-intensive Visual Reasoning Abstract Vision-Language Models (VLMs) should answer from concrete image evidence rather than language priors, dataset shortcuts, or irrelevant visual context. Existing perception-aware post-training methods encourage image use through global perturbations or attention proxies, but they do not test whether a sampled answer causally depends on the local evidence that supports it. We propose Counterfactual Evidence Disentanglement (CED), a t