知识卡片
表面相关可能放大社会偏差
训练数据中外观属性与职业或活动相关,模型可能把这些统计捷径当作预测依据,并在输出中延续或放大偏差。需要区分任务所需信息与不恰当代理。删除一个显式字段不保证图像里不存在其它相关线索。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 计算机视觉与社会
原书出处 · 1 处
- · 查看原书依据
Correlations between the expressed gender of individuals and their depicted activities recorded within a dataset can perpetuate or even amplify societal gender biases captured in the dataset [527, 92, 356]. Gender classification systems in 2018 showed performance that depended on skin tone [65].
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:摘录中关于表面相关可能放大社会偏差的定义、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:Correlations between the expressed gender of individuals and their depicted activities recorded within a dataset can perpetuate or even amplify societal gender biases captured in the dataset [527, 92, 356]. Gender classification systems in 2018 showed performance that depended on skin tone [65].