CLIMB
收藏资源简介:
CLIMB数据集由南加州大学、加州大学洛杉矶分校和加州大学戴维斯分校共同创建,旨在评估大型语言模型在临床决策中的内在和外在偏见。该数据集包含94739条临床诊断数据,来源于MIMIC-IV数据库中的去标识化电子健康记录。数据集的创建过程结合了ICD-10-CM代码和人口统计信息,通过引入新的评估指标AssocMAD来量化模型在不同人口统计群体中的表现差异。CLIMB数据集主要应用于医疗领域的偏见评估,旨在提高临床决策的公平性和准确性。
The CLIMB dataset was co-developed by the University of Southern California, University of California, Los Angeles, and University of California, Davis, aiming to evaluate the intrinsic and extrinsic biases of large language models in clinical decision-making. This dataset comprises 94,739 clinical diagnostic records sourced from de-identified electronic health records (EHRs) in the MIMIC-IV database. Its construction process integrates ICD-10-CM codes and demographic information, and introduces a novel evaluation metric, AssocMAD, to quantify the performance disparities of models across different demographic groups. The CLIMB dataset is primarily applied for bias assessment in the healthcare field, with the goal of enhancing the fairness and accuracy of clinical decision-making.
CLIMB: A Benchmark of Clinical Bias in Large Language Models
数据集概述
- 名称: CLIMB
- 全称: A Benchmark of Clinical Bias in Large Language Models
发布计划
- 代码和数据: 即将发布

- 1CLIMB: A Benchmark of Clinical Bias in Large Language Models南加州大学;加州大学洛杉矶分校;加州大学戴维斯分校 · 2024年



