Re: [UAI] Belief functions and continuous domains

From: Russell Almond (ralmond25@home.com)
Date: Sun Apr 02 2000 - 15:24:36 PDT

  • Next message: Tim Wilkin: "[UAI] DBN inference..."

    "Kevin S. Van Horn" wrote:

    > Can anyone tell me if belief-function theories such as Dempster-Shafer are
    > capable of handling continuous domains? Shafer's book only discusses finite
    > domains, and it's not at all obvious to me how one would generalize it to a
    > continuous domain.

    Yes. In fact the oft cited seldom read Dempster [1968] paper develops the
    belief function generalization of the beta distribution.

    The problem you are having is that if your domain is the real line between 0
    and 1 (size aleph 1) then the domain of the measure function m() whould be of
    size alph 2. The same thing happens in probability because P() is a set
    function. You just pick a Borel set over the domain (i.e., the set of all
    intervals plus their unions).

    The useful belief functions tend to be random intervals whose endpoints have a
    probability distribution of a known form. Dempster develops a "bivariate beta"
    distribution in his paper. I develop a bivariate gamma in my book (Graphical
    Belief Modeling).

        --Russell Almond



    This archive was generated by hypermail 2b29 : Sun Apr 02 2000 - 15:28:30 PDT