Given the Bayesian network on the left hand side in the following figure, it shows that the random variable $B$ is dependent on $A$ and $C$, and the Bayesian network $G$ can be factorized as: $P(G) = p(A) \times p(C) \times p(B|A,C)$. The Bayesian network $G$ can be moralized to a Markov network $M$ for junction tree algorithm and it is shown on the right hand side of the figure. In $M$, all the existing edges in $G$ are changed to undirected ones and it adds a new undirected edge between the parents of $B$. Is it correct to use the same probability $P(G) = p(A) \times p(C) \times p(B|A,C)$ to factorize $M$ (assuming the factors over $A$, $B$, and $C$ are probabilities)? I understand that a new edge is added between $A$ and $C$, but I think the edge is added to capture the dependencies in the Markov network, is it compulsory to introduce an edge factor/potential in the factorization of $M$? enter image description here


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.