Talk:Markov blanket

From Wikipedia, the free encyclopedia

There's also a more general meaning to "Markov blanket" used in at least one book on pattern recognition, or perhaps "additional use". When removing redundant variables one may start out with all variables, removing one at a time. Then the markov blanket of a variable, far as I recall, is a set of variables that make the given variable redundant. If the blanket is found, it can then be safely removed and it won't be needed after future iterations.

[edit] Co-parents

I would like to suggest that "children's parents" be replaced by the term "co-parents".

I think that this is technically more accurate since the term "co-parents" excludes the concerned node A that we wish to separate from the graph while "children's parents" involves also A.

Secondly, it is a single word and perhaps easier to remember and helps visualise the relationship of node A with the concerned nodes. Gugux (talk) 18:03, 6 March 2008 (UTC)

[edit] Conditional Independence

I changed \Pr(A \mid \partial A \cap B) = \Pr(A \mid \partial A) to \Pr(A \mid \partial A , B) = \Pr(A \mid \partial A) because the \cap sign means intersection, and the intersection \partial A \cap B only includes the node B. Instead we want the union of \partial A and B, which would be \partial A \cup B, but I believe the comma is more standard notation. Sunbeam44 (talk) 22:31, 11 April 2008 (UTC)

I just realized I was confusing sets of nodes with probability notation. Union would also be incorrect. However, using the intersection notation on variables is confusing-- this notation is typically used for an intersection of events. Since here we have variables and not events, the short-hand notation of comma is better in my opinion. Sunbeam44 (talk) 18:07, 23 May 2008 (UTC)