← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 6598.c · 生成模型:GPT-6

部分可观测状态须维护信念

理解等级:高级

POMDP中观测只是隐藏状态的带噪信息,当前观测不一定足以预测未来。可以依据行动与观测历史维护状态信念,并据此选择行动;隐藏状态图和观察机制应明确,而非把观测变量直接改名为完整状态。

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

      12.5.3 Partially observable MDPs An extension of MDPs is partially observed MDPs (POMDPs). In a POMDP, the agent doesn’t know with certainty what state they are in, and they must make inferences about that state given incomplete evidence from their environment. This applies to many practical problems where the agent cannot observe the full state of the environment. A POMDP can entail different causal structures

    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:摘录中关于部分可观测状态须维护信念的定义、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:12.5.3 Partially observable MDPs An extension of MDPs is partially observed MDPs (POMDPs). In a POMDP, the agent doesn’t know with certainty what state they are in, and they must make inferences about that state given incomplete evidence from their environment. This applies to many practical problems where the agent cannot observe the full state of the environment. A POMDP can entail different causal structures