Coupled or Independent? Fault Coupling in Agentic Medical Vision Systems

查看原文 / View Original 编辑 / Edit

Computer Science > Computer Vision and Pattern Recognition arXiv:2610.40913 (cs) [Submitted on 29 Oct 2026] Title: Coupled or Independent? Fault Coupling in Agentic Medical Vision Systems Authors: Nishi Kanta Paul, Md Shihabul Islam Shovo, Md Mostafijur Rahman, Jamil Fayyad.

计算机科学 > 计算机视觉与模式识别 arXiv:2610.40913 (cs) [提交于 2026 年 10 月 29 日] 标题：耦合还是独立？代理医疗视觉系统中的故障耦合作者：Nishi Kanta Paul, Md Shihabul Islam Shovo, Md Mostafijur Rahman, Jamil Fayyad。

Abstract: Vision-language agents equipped with tools expose intermediate outputs for inspection and comparison with the model’s reasoning, which may offer safety benefits over single-shot models. This paper tests that potential benefit when the sensor reading itself is untrusted, so that the tool and reasoning channels share a common visual fault.

摘要：配备工具的视觉语言代理会公开中间输出，以便进行检查并与模型的推理进行比较，这可能比单次触发模型提供更高的安全性。本文在传感器读数本身不可信的情况下测试了这种潜在优势，即工具通道和推理通道共享一个共同的视觉故障。

We consider a dermatology agent based on segmentation, contrastive retrieval, and an iterative vision-language reasoner and define a coupling taxonomy that distinguishes single-channel errors from joint errors that agree or disagree. On PAD-UFES-20, we show that typographic injection produces detectable directional retrieval capture, whereas physical corruption does not.

我们考虑了一种基于分割、对比检索和迭代视觉语言推理器的皮肤科代理，并定义了一种耦合分类法，用于区分单通道错误与一致或不一致的联合错误。在 PAD-UFES-20 数据集上，我们证明了印刷体注入会产生可检测的方向性检索捕获，而物理损坏则不会。

Crop-first retrieval preserves the payload at a readable scale and produces 55% joint wrong agreement across all cases (71% of faulty cases); full-image retrieval produces 35% (43%). Under a binary decision, if both channels are wrong, their predictions must agree, so hard-label channel comparison cannot detect joint failures, limiting disagreement recall to 0.29 (crop-first) versus 0.57 (full-image). The direction of the order effect is reproduced on the independent HIBA test set.

“先裁剪后检索”方式在可读尺度上保留了有效载荷，并在所有案例中产生了 55% 的联合错误一致性（在故障案例中占 71%）；全图检索则产生 35%（43%）。在二元决策下，如果两个通道都出错，它们的预测必然一致，因此硬标签通道比较无法检测到联合故障，将不一致召回率限制在 0.29（先裁剪）对比 0.57（全图）。这种顺序效应的方向在独立的 HIBA 测试集上得到了重现。