小辣机学数学系列
在可信的贝叶斯网络中,一个节点的马尔可夫毯包括它的父节点、子节点和子节点的父节点。
例子: A A A和 B B B是否关于 F F F独立?
P ( A , B ∣ F ) = ∑ E ∑ C P ( A , B , C , E ∣ F ) = ∑ E ∑ C P ( A ) ⋅ P ( E ∣ A , F ) ⋅ P ( B ∣ F ) ⋅ P ( C ∣ E ) = P ( A ) ⋅ P ( B ∣ F ) ∑ E ∑ C [ P ( E ∣ A , F ) ⋅ P ( C ∣ E ) ] = P ( A ) ⋅ P ( B ∣ F ) ∑ E [ P ( E ∣ A , F ) ∑ C P ( C ∣ E ) ] = P ( A ) ⋅ P ( B ∣ F ) ∑ E [ P ( E ∣ A , F ) ] = P ( A ) ⋅ P ( B ∣ F ) = P ( A ∣ F ) ⋅ P ( B ∣ F ) \begin{aligned} P(A,B|F)&=\sum_E\sum_CP(A,B,C,E|F)\\ &=\sum_E\sum_CP(A)\cdot P(E|A,F)\cdot P(B|F)\cdot P(C|E)\\ &=P(A)\cdot P(B|F)\sum_E\sum_C\left[P(E|A,F)\cdot P(C|E)\right]\\ &=P(A)\cdot P(B|F)\sum_E\left[P(E|A,F)\sum_CP(C|E)\right]\\ &=P(A)\cdot P(B|F)\sum_E\left[P(E|A,F)\right]\\ &=P(A)\cdot P(B|F)\\ &=P(A|F)\cdot P(B|F) \end{aligned} P(A,B∣F)=E∑C∑P(A,B,C,E∣F)=E∑C∑P(A)⋅P(E∣A,F)⋅P(B∣F)⋅P(C∣E)=P(A)⋅P(B∣F)E∑C∑[P(E∣A,F)⋅P(C∣E)]=P(A)⋅P(B∣F)E∑[P(E∣A,F)C∑P(C∣E)]=P(A)⋅P(B∣F)E∑[P(E∣A,F)]=P(A)⋅P(B∣F)=P(A∣F)⋅P(B∣F)
所以 A A A和 B B B关于条件 F F F独立。
定理:
- 给定变量X的父节点,则X独立于非X后代的所有节点
- 给定变量X的马尔科夫毯,则X独立于其它变量