2
$\begingroup$

If A and B don't have parents on a Bayesian network, does that mean we can infer A and B are independent? Whatever the answer is, please explain how you get to it.

$\endgroup$
4
$\begingroup$

Yes, they would be unconditionally independent.

This is simply a consequence of the factorisation of a Bayesian network into conditional probabilities. For example, this network will factorise as $p(A,B,C)=p(C|A,B)p(A)p(B)$, implying that $p(A,B)=p(A)p(B)$, which is the very definition of unconditional independence for the parent-less nodes $A$ & $B$:

enter image description here

As for the factorisation $p(A,B,C)=p(C|A,B)p(A)p(B)$ that led us here, this is definitional for Bayesian networks. I recommend Section 3.3 of Barber for a fuller explanation.

On the other hand, they are conditionally dependent if conditioned on a collider. For example, nodes $x$ and $y$ are graphically conditionally dependent when conditioning on the collider $z$ in each of these 2 BNs:

enter image description here

Chapter 3 of the same source contains a fuller description of BNs and similar graphical networks, together with various possible tests for conditional independence.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.