[UAI] A general-purpose algorithm for belief networks?

From: Briggs, Will (Briggs_W@mail.lynchburg.edu)
Date: Tue Apr 03 2001 - 13:47:22 PDT

  • Next message: Fabio Gagliardi Cozman: "Re: [UAI] A general-purpose algorithm for belief networks?"

    Hi,

    I have a student that's planning to do an anytime belief network. (There
    was mail about this earlier; thanks for the references!)

    For this, we need an algorithm that handles belief nets. I find in
    _Probabilistic Reasoning_ and in Russell & Norvig some algorithms for belief
    nets that are polytrees, or using some particular tweak to get around the
    intractability.

    Is there an algorithm that would handle any belief net?
    Has anyone written it down or implemented it?

    I had thought of using the one Russell & Norvig presents for polytrees, on
    nets that _aren't_ polytrees. (I do realize this will give the wrong
    answer, but hope we can tweak it to fix the problem.) Question: there is a
    probability

    P(Causes for X|Effects of X) //X is the variable we're evaluating

    which is called a normalizing constant, and I don't know how to derive it.
    Normalization as in gamma distribution? Uniform distribution? Some other
    kind of normalization?

    Thanks for any and all,
    WSB

    This archive was generated by hypermail 2b29 : Tue Apr 03 2001 - 13:52:32 PDT