arXiv cs.AI· Chunzheng Zhu, Jiaqi Zeng, Hongbo Zhao, Yihang Chen, Yijun Wang, Jianxin Lin·· 5 小时前AI 评分37
CRAFT：医疗视觉语言模型的因果责任与失败追踪框架
CRAFT: Causal Responsibility and Failure Tracing in Medical Vision Language Models
AI 导读
医疗 VLM 存在两类安全失败：文本上下文压过图像判断的仲裁失败，以及证据不足仍自信作答的刹车失败。CRAFT 通过双重准则将两种失败各自定位于空间上不重叠的注意力头集合——仲裁头分布在中到深层宽频带，刹车头集中在中到晚层窄频带——并用时间探针和 Tuned Lens 轨迹分析验证必要性。
正文 · 原文
Abstract:As vision language models are increasingly deployed in clinical diagnosis, understanding how they internally resolve competing visual and textual signals becomes a safety imperative. Existing mechanistic analyses remain confined to unimodal text and offer no explanation for why a single misleading sentence can override a correct image based diagnosis, or why a model commits to a confident answer despite insufficient visual evidence. We find that these two safety risks, arbitration failure where textual context overrides visual grounding and brake failure where the model commits without adequate evidence, are mediated by spatially disjoint attention head populations: arbitration heads form a mid-to-deep wideband reflecting cross-layer evidence competition, while brake heads concentrate in a narrow middle-to-late layer band that regulates evidence sufficiency and abstention behavior. To ground these observations in causal circuitry, we introduce CRAFT, which localizes each failure mode to a minimal causal head set via dual criteria and verifies necessity and sufficiency through temporal probes and Tuned Lens trajectory analysis. Excising arbitration heads sharply reduces conflict following with negligible degradation on clean inputs, while excising brake heads restores appropriate abstention under degraded visual evidence. The two interventions target spatially disjoint head sets and produce distinct corrective effects, underscoring the mechanistic separability of the failure modes. Experiments across multiple medical VQA benchmarks and VLM architectures validate both the localization and interventions, demonstrating that the identified heads causally drive each failure mode and that targeted modulation generalises without retraining. The code is available at this https URL.
| Comments: | NeurIPS 2026 Spotlight, Medical VLM Failure Analysis |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.38810 [cs.CV] |
| (or arXiv:2609.38810v3 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2609.38810 arXiv-issued DOI via DataCite |
Submission history
From: Jiaqi Zeng [view email]
[v1]
Wed, 30 Sep 2026 02:41:33 UTC (4,678 KB)
[v2]
Fri, 2 Oct 2026 15:11:54 UTC (4,678 KB)
[v3]
Thu, 29 Oct 2026 10:40:34 UTC (4,680 KB)
来源：arXiv cs.AI · arxiv.org