Derivation of P(MB(X)) where MB(X) is the Markov Blanket of X in a Bayesian Network

482 Views Asked by At

Given the Markov Blanket $\mathit MB(X)$ I am told that $$P(\mathit MB(X)) = \alpha P(X \vert U_{1}, \cdots , U_{n}) \prod_{Y_{i}} P(Y_{i} \vert P(Y_{i} \vert Z_{i1} \cdots)$$ where $\alpha$ is the normalization factor and $\mathit MB(X)$ is represented by this diagram reproduced from Artificial Intelligence A Modern Approach 3rd Edtion .From Artificial Intelligence A Modern Approach 3rd Ed.

Thus far I've been unable to figure out how this equation is derived. An approach I've tried is applying the exact inference equation and the definition of a joint distribution for Bayesian Networks: $$P(X\vert e) = \alpha P(X, \text{e}) = \alpha \sum_{y \in \pmb Y}P(X, \text{e}, \text{y})$$ $$P(x_{1}, \cdots ,x_{k}) = \prod_{i=1}^{k}P(x_{i} \vert \textit{parents}(X_{i}))$$ where $\pmb E$ is the is the set of evidence variables and $\text{e} \subseteq \pmb E$ are the evidence variables, $\pmb Y$ is the set of hidden variables and all the variables in the network is $\pmb X = \{X\} \cup \pmb E \cup \pmb Y$, and $\textit{parents}(X_{i})$ are the parent nodes of the node corresponding to the RV $X_{i}$ in the Bayesian Network.

The issue I'm facing is that when I expand the Markov Blanket probability, I end up with factors depending on the parent's of the $Z_{ij}$ nodes, which do not appear in the LHS of the equation I was given. Can someone provide me with the correct derivation of the RHS and give me some detail on the approach used?