Re: ?Value of information without utilities?

[Rina Dechter]

>Does anyone know of methods for representing constraints in a Bayesian
>network?
>

There is the staightforward approach of expressing every constraint
as a CPT (variables are the parents and the constraint
value (true-false) as a child.) Then you can use evidence to force
the child to true. This appears at Pearl's book and in Coopers
Proofs of Np-hardness. However, this does not facilitate
constraint propagation which seem to be the reason
one would want to have a special representation for constraints.
For some preliminary work in that direction.
see the technical report R67 in 
www.ics.uci.edu/~dechter/publication1.html

-----Rina.
>Bob Welch
>Gensym Corporation
>
>