RL微调VLM的鲁棒性与思维链一致性研究

On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs

AI 导读

强化学习（RL）微调被扩展至视觉语言模型（VLM）。研究发现，简单的文本扰动——误导性标题或错误思维链（CoT）——会显著降低模型鲁棒性和置信度，且开源模型衰退更明显。闭源模型呈现类似失败模式，但鲁棒性和推理一致性更强。进一步分析揭示准确性与忠实性的权衡：微调提升基准准确率，但同时侵蚀CoT的可靠性及对上下文变化的鲁棒性；对抗性增强可改善鲁棒性，却无法阻止忠实性漂移。引入忠实性感知奖励能恢复答案与推理的对齐，但与增强结合时训练易崩溃到捷径策略。这些发现强调需联合关注正确性、鲁棒性与视觉推理的忠实性。

推荐理由

RL微调让VLM基准分变好看，却可能让它的推理链变得靠不住，这个反直觉的诊断对正在用RL打磨多模态模型的团队是个警醒。

正文 · 原文

Reinforcement learning (RL) finetuning has become a key technique for enhancing large language models (LLMs) on reasoning-intensive tasks, motivating its extension to vision language models (VLMs). While RL-tuned VLMs improve on visual reasoning benchmarks, they remain vulnerable to weak visual grounding, hallucinations, and over-reliance on textual cues. We show that simple, controlled textual perturbations—misleading captions or incorrect chain-of-thought (CoT) traces—cause substantial drops in robustness and confidence, and that these effects are more pronounced when CoT consistency is taken into account across open-source multimodal reasoning models. In contrast, closed models exhibit similar failure modes but maintain markedly greater robustness and reasoning consistency, suggesting that the gap reflects a shortcoming in current open-source RL finetuning rather than an inherent limitation of the task. To better understand these vulnerabilities, we further analyze RL finetuning dynamics and uncover an accuracy–faithfulness trade-off: finetuning raises benchmark accuracy, but can simultaneously erode the reliability of the accompanying CoT and its robustness to contextual shifts. Although adversarial augmentation improves robustness, it does not by itself prevent faithfulness drift. Incorporating a faithfulness-aware reward can restore alignment between answers and reasoning, but when paired with augmentation, training risks collapsing onto shortcut strategies and robustness remains elusive. Together, these findings highlight the limitations of accuracy-only evaluations and motivate training and assessment protocols that jointly emphasize correctness, robustness, and the faithfulness of visually grounded reasoning.

† Harvard University
‡ OpenAI
** Work done while at Apple