1
$\begingroup$

I've been teaching myself about d-separation and am trying to answer the following question. Given the graphs below, write down all conditional independence relationships involving the random variable C i.e. $C \perp X|Y$

Note: I've had to use $\perp$ for conditional independence as I don't know how to write the proper symbol in LaTeX without adding stuff to preamble (which stackexchange doesn't have).

Here are the graphs:

enter image description here

For image (i), we have a tail-to-tail relationship at B which means C and A are d-separated and so I have $C \perp A|B$. I cannot find any more conditional independence relationships here of the form $C \perp X | Y$ since both $B$ and $D$ are adjacent to $C$ and therefore not valid. Is this right?

For image (ii), we have a head-to-tail relationship at B and D. I think this means $C \perp A|B$ and $C \perp A|D$ and $C \perp A| \{ B,D\}$. I was tempted to write $C \perp B|D$ but I don't think this is valid since only one of the possible paths from C to B would be d-separated. So I find a total of three valid conditional indpendence relationships. Is that right?

For image (iii) and (iv), the problem is now that the graphs are undirected and I have not been able to find as much material online to help with this. I was wondering if someone could help with the following:

(a) explain the meaning of the black squares in (iv) (b) link to some notes/videos that might help with finding conditional independencies in undirected graphs

Thanks!

$\endgroup$
1
$\begingroup$

(a) The black squares represent factors in a factor graph. Wikipedia is the wiki page about factor graphs.

(b) A pdf from stat cmu is a reference about Undirected Graph / Markov Random Field.

$\endgroup$
  • $\begingroup$ Thanks. Yeah I found some helpful stuff as well. Understand it all now :) $\endgroup$ – user11128 Sep 27 at 8:03

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.