知识卡片

普通读书笔记卡 · 6595.g · 生成模型:GPT-6

贝叶斯可容许保证需要条件

理解等级:高级

有限参数空间、有限损失且先验全支持时,Bayes规则有可容许性保证;更一般空间需额外条件。近似推断与近似优化还可能偏离精确Bayes规则,不能只因使用Pyro或给参数加先验就宣称自动获得保证。

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

      mild regularity conditions. This means that if we use Bayesian inference in Pyro or similar libraries to calculate and optimize posterior expected loss, we have an admissibility guarantee (if those mild conditions hold, and they usually do). That means you needn’t worry that someone else’s decision-making model (that makes the same modeling assumptions, has the same utility function, and uses the same data) will beat

    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:摘录中关于贝叶斯可容许保证需要条件的定义、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:mild regularity conditions. This means that if we use Bayesian inference in Pyro or similar libraries to calculate and optimize posterior expected loss, we have an admissibility guarantee (if those mild conditions hold, and they usually do). That means you needn’t worry that someone else’s decision-making model (that makes the same modeling assumptions, has the same utility function, and uses the same data) will beat