← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6574.b · 生成模型:GPT-6

只改假想世界保留事实证据

理解等级:中级

查询P(Yx′|X=x,Y=y)时,事实世界仍保留X=x、Y=y的机制和证据;只在假想世界替换X的赋值。若把事实图也一并do掉,可能丢失事实对外生原因的约束,进而改变所求反事实。

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

      8.4.3 Applying the hypothetical condition via graph surgery The hypothetical world will, typically, differ from the actual world by the hypothetical condition. For example, in P (IE=“high”|E = “low”, I < 50), “if engagement were high” (E=“high”) differs from the factual condition “engagement was low” (E = “low”). As we’ve discussed, we model the hypothetical condition with the ideal intervention—we intervene on E, setting it

    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:摘录中关于只改假想世界保留事实证据的定义、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:8.4.3 Applying the hypothetical condition via graph surgery The hypothetical world will, typically, differ from the actual world by the hypothetical condition. For example, in P (IE=“high”|E = “low”, I < 50), “if engagement were high” (E=“high”) differs from the factual condition “engagement was low” (E = “low”). As we’ve discussed, we model the hypothetical condition with the ideal intervention—we intervene on E, setting it