知识卡片
观测王子需要更新国王分布
在K→P、K→Y、P→Y图中,观察P=p时需按p(K|P=p)平均p(Y|K,p);干预P=p则按原p(K)平均。两者可能不同,根源是观察提供了K的信息,而干预切断其对P的赋值影响。
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
- · 因果与大语言模型
原书出处 · 1 处
- · 查看原书依据
By now, you know the distribution P(Kingdom|Prince=x) will be different from P(KingdomPrince=x), but let’s demonstrate the fact with this causal LLM. First, we’ll model P(Kingdom|Prince=x), where x is His courageous Prince takes command, leading the kingdom’s army to victory in battle after battle To infer P(Kingdom|Prince=x), we’ll have to infer the distribution of the latent confounder, King. We’ll do this using the prince2king_model we trained.
参考来源
- 位置:《Causal AI》第13章《因果与大语言模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-13.xhtml)
- 结论依据:摘录中关于观测王子需要更新国王分布的定义、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:By now, you know the distribution P(Kingdom|Prince=x) will be different from P(KingdomPrince=x), but let’s demonstrate the fact with this causal LLM. First, we’ll model P(Kingdom|Prince=x), where x is His courageous Prince takes command, leading the kingdom’s army to victory in battle after battle To infer P(Kingdom|Prince=x), we’ll have to infer the distribution of the latent confounder, King. We’ll do this using the prince2king_model we trained.