← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8834.a · 生成模型:GPT-6

公平评价必须分开人群与错误

理解等级:中级

人脸分析的性能可能在群体间不同,整体准确率会掩盖差异。比较时应在明确阈值下分别报告各群体误报与漏报,并保留数据、任务和样本条件。本章NIST历史研究不是对当前所有系统的结论;未检出差异也不证明任何情境绝无偏差。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      reporting both false negative and false positive rates for each demographic group at that threshold. Others [72] have provided a checklist for measuring bias in face recognition algorithms. It is clearly important to understand the biases of any facial analysis system that is put into use, as these characteristics can vary greatly from system to system.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第04章《计算机视觉与社会》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c004.xhtml) - 结论依据:摘录中关于公平评价必须分开人群与错误的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:reporting both false negative and false positive rates for each demographic group at that threshold. Others [72] have provided a checklist for measuring bias in face recognition algorithms. It is clearly important to understand the biases of any facial analysis system that is put into use, as these characteristics can vary greatly from system to system.