知识卡片
响应核不能唯一决定反事实轨迹
多个结构方程可产生同一p(S_next|S,A),却把不同潜在行动结果按不同方式耦合。要模拟同一次遭遇下改行动的轨迹,需要额外指定并论证外生噪声与机制;仅知道转移概率或能随机采样还不够。
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
原书出处 · 1 处
- · 查看原书依据
P (S t +1 = s'|St = s, At = a) captures the rules of state changes in the environment. As I mentioned, P (S t +1|St = s, At = a) is the causal Markov kernel for a given state S t +1. We could specify an SCM with an assignment function that entails that causal Markov kernel, and write that assignment function as
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:摘录中关于响应核不能唯一决定反事实轨迹的定义、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:P (S t +1 = s'|St = s, At = a) captures the rules of state changes in the environment. As I mentioned, P (S t +1|St = s, At = a) is the causal Markov kernel for a given state S t +1. We could specify an SCM with an assignment function that entails that causal Markov kernel, and write that assignment function as