d-separation results from the structure of Bayes Nets; with all the
probabilistic information stored in the (input) conditional probability
tables of the nodes, d-seperation can be proved. The justification for
storing the probability information in just the node conditional
probability tables usually derives from the what has been termed the Causal
Markov Condition. Many pople take this condition to be the essence, perhaps
even the definition of causality.
For a contrary view and a fuller explication of the Causal Markov
condition, see my paper "The Causal Markov Condition, Fact or Artifact?" in
SIGART, Vol 7 #3. If you don't have access to this paper, I will be more
than happy to send you a copy.
John
>Hello UAI People!
>
> I woud like to thanks to anyone that help me, and
>to make more questions about bayesian belief networks.
> I'm a inexperienced student in AI, and I would like
>to know something.
> Is D-separation a criterion to avoid unusefull computation
>in belief networks? Sorry for this student question!
>
>Thanks
>
>
>José Carlos
John F. Lemmer, PhD
AFRL/IFTB
525 Brooks Road
Rome, NY 13441
(315)330 3657
DSN 587 3657