发表机构
Politecnico di Milano; Zhejiang University; China-Japan Friendship Hospital; Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences; Emory University（米兰理工大学; 浙江大学; 中日友好医院; 中国科学院深圳先进技术研究院; 埃默里大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究定义了派生特征过度信任（DFOT）问题，构建了量化该问题的估计量框架，在PPG-ECG数据集上验证了基于特权蒸馏的基线方法可缓解DFOT，为相关研究提供通用评估目标。
AI 中文摘要
派生测量值作为直接事实越来越多地进入大语言模型（LLM）的流程中，尽管它们的有效性依赖于具体实例。我们将派生特征过度信任（DFOT）定义为下游LLM的一种失效情况：它将此类测量值赋予直接事实的认知地位，或在其有效范围之外使用该测量值。以生理传感为例，D1测试LLM接受与离线心电图（ECG）矛盾的光体积描记法（PPG）派生节律的情况，D2测试LLM在存在误导性严重历史的情况下拒绝经离线确认的可靠PPG节律的情况。ECG提供训练监督和离线参考构建，但从不提供给LLM。五个估计量对该问题链进行量化：冲突过度信任率（COTR）和情境诱导错误率（CIR）分别表征D1和D2；正确修复率（CRR）衡量冻结错误的修复情况；特定证据修复边际（ESRM）对比匹配证据与患者不相交的打乱证据；效用损害率（UHR）衡量基线中无需验证的高可靠性案例中出现的不必要验证情况。该框架不依赖于特定的可靠性生成器，我们使用50000对PPG-ECG记录，以ECG-to-PPG特权蒸馏作为说明性基线，在仅使用PPG的推理场景中对其进行验证。在协议锁定的187名患者的测试集上，基线方法使四个修复和特异性指标提升了1.82至6.69个百分点，所有配对置信区间均不包含零；UHR上升了0.67个百分点（95%置信区间：-0.4至+1.7）。DFOT为更强的缓解方法提供了通用评估目标，代码可在指定URL获取。
英文摘要
Derived measurements increasingly enter large language model (LLM) pipelines as direct facts despite their instance-dependent validity. We define derived-feature over-trust (DFOT) as the failure in which a downstream LLM assigns such a measurement the epistemic status of a direct fact or uses it outside its valid scope. Using physiological sensing as a case study, D1 tests acceptance of a PPG-derived rhythm contradicted by offline ECG, whereas D2 tests rejection of an offline-confirmed reliable PPG rhythm under misleading severe history. ECG supplies training supervision and offline reference construction but is never shown to the LLM. Five estimands quantify this chain: conflict over-trust rate (COTR) and context-induced error rate (CIR) characterize D1/D2; correct repair rate (CRR) measures frozen-error repair; evidence-specific repair margin (ESRM) contrasts matched and patient-disjoint shuffled evidence; and utility harm rate (UHR) measures unnecessary verification among HIGH-reliability cases used without verification at baseline. The framework does not depend on a particular reliability generator. We demonstrate it on 50,000 paired PPG-ECG records using ECG-to-PPG privileged distillation as an illustrative baseline and PPG-only inference. On a protocol-locked 187-patient test, the baseline improves four repair and specificity endpoints by 1.82-6.69 percentage points, with all paired confidence intervals excluding zero; UHR increases by 0.67 percentage points (95% CI: -0.4 to +1.7). DFOT provides a common evaluation target for stronger mitigation methods. The code is available at https://github.com/Zongheng-Guo/When-Derived-Measurements-Mislead.
Comments25 pages, including references and supplementary material; 3 figures and 19 tables. Code: https://github.com/Zongheng-Guo/When-Derived-Measurements-Mislead
