← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6595.b · 生成模型:GPT-6

期望效用对干预结果积分

理解等级:中级

若效用为U(y),行动x的期望价值为Σ_y U(y)P(Y_x=y),连续情形相应积分。选择其最大值须先得到适合该行动的干预结果分布;期望收益最大也不等于每次实现收益最大。

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

      Maximizing expected utility The most intuitive and commonly seen decision rule is to choose the action that maximizes expected utility. First, we can look at the expectation of the utility distribution. Since utility is a deterministic function of YX=x, this is just the expectation of U (YX=x) over the intervention distribution of Y. We then choose the action (value of x) that maximizes expected utility:

    参考来源

    - 位置:《Causal AI》第12章《因果决策与强化学习》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-12.xhtml) - 结论依据:摘录中关于期望效用对干预结果积分的定义、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:Maximizing expected utility The most intuitive and commonly seen decision rule is to choose the action that maximizes expected utility. First, we can look at the expectation of the utility distribution. Since utility is a deterministic function of YX=x, this is just the expectation of U (YX=x) over the intervention distribution of Y. We then choose the action (value of x) that maximizes expected utility: