知识卡片
行动补救需加入可实施约束
算法补救希望给出当事人可采取且有望改变决策的行动,因此应加入可变性、成本、依赖关系和执行限制。某个可到达特征向量在当前分类器中被批准,只是一项模型条件结论,不是现实批准或收益保证。
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
- · 反事实与平行世界
原书出处 · 1 处
- · 查看原书依据
Increasing salary by $20,000 is unrealistic for most loan applicants. That’s where counterfactual-based algorithmic recourse can be useful. Algorithmic recourse looks for the nearest hypothetical condition that would have led to a different classification. It operates under the constraint that the hypothetical condition was achievable or actionable by the applicant. Figure 8.2 shows how this works.
参考来源
- 位置:《Causal AI》第08章《反事实与平行世界》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-8.xhtml)
- 结论依据:摘录中关于行动补救需加入可实施约束的定义、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:Increasing salary by $20,000 is unrealistic for most loan applicants. That’s where counterfactual-based algorithmic recourse can be useful. Algorithmic recourse looks for the nearest hypothetical condition that would have led to a different classification. It operates under the constraint that the hypothetical condition was achievable or actionable by the applicant. Figure 8.2 shows how this works.