[论文] Making Clinical Language Models Auditable: Concept-Guided Fine-Tuning ...

论文概要

研究领域: NLP
作者: Jin Mu, Guanhua Chen
发布时间: 2026-08-27
arXiv: 2608.27397

中文摘要

临床语言模型可以在院内实现强准确率，但在部署转换下失败，因为它们利用不反映患者状态的笔记特定伪影（如模板、分隔符、样板）。我们提出了CAST（概念引导的伪影抑制调优），一个基于SAE的可审计临床文本分类框架。CAST使用稀疏自编码器从中间Transformer激活中暴露稀疏、人类可审计的特征，用LLM辅助解释管道和ICD-10检索约束标记SAE潜在变量，通过残差减法在微调期间抑制已验证的伪影潜在变量，并为审计模型决策提供事后每概念归因。在MIMIC-IV出院笔记死亡率预测上，CAST优于相应的微调编码器基线，与强LLM基线保持竞争力，同时产生支持每次预测的临床概念和训练期间抑制的伪影概念的特征级审计追踪。

原文摘要

Clinical language models can achieve strong in-hospital accuracy yet fail under deployment shifts because they exploit note-specific artifacts (e.g., templates, separators, boilerplate) that do not reflect patient state. We propose CAST (Concept-guided Artifact Suppression Tuning), an SAE-based framework for auditable clinical text classification. CAST uses Sparse Autoencoders to expose sparse, human-auditable features from intermediate Transformer activations, labels SAE latents with an LLM-assisted interpretation pipeline and ICD-10 retrieval constraints, suppresses verified artifact latents via residual subtraction during fine-tuning, and provides post-hoc per-concept attributions for auditing model decisions. On MIMIC-IV discharge-note mortality prediction, CAST improves over its corre...

自动采集于 2026-08-30

#论文 #arXiv #NLP #小凯