J. Pearl devoted an entire chapter (Ch 9) to a comparison of Bayesian
versus Dempster Shafer.
Judea Pearl, Probabilistic Reasoning in Intelligent Systems, Morgan
Kaufmann, 1986.
In making a comparison I would suggest you include a comparison with a
Bayesian network that is built from causal relationships (and so exhibits
lots of conditional independence and absence of directed arcs) where causes
are the roots versus evidence based networks where the symptoms are the
roots, which is the Dempster Shafer framework as a "Theory of Evidence". An
evidence based system has problems sorting out the natural dependence among
symptoms, where as the causal based Bayes net is very good in anticipating
that dependence.
Also, you may want to consider development costs. In assessing the evidence
based system, the developer has to be careful to account for the dependence
among symptoms.
Bob Welch
Gensym Corporation
4940 Pearl East Circle
Boulder Colorado 80303
(303) 440-0400
bwelch@gensym.com
-----Original Message-----
From: Antonio Salmeron <asc@stat.ualm.es>
To: Jim Myers <myers29@erols.com>
Cc: uai@CS.ORST.EDU <uai@CS.ORST.EDU>
Date: Friday, May 14, 1999 6:15 AM
Subject: Re: Dempster-Shafer vs Bayes
Jim Myers wrote:
> Dear Colleagues
>
> I would appreciate any pointers to papers comparing the Dempster-Shafer
and
> Bayesian approaches. I'm looking for papers that cover both theory and,
> especially, applications. I have a copy of David Heckerman's 90 UAI
paper.
>
> Also, I would appreciate any ideas on how to best compare the two methods
> empirically. How should I design experiments, what metrics, any other
> ideas?
>
> Thank you in advance
> Best Regards
> Jim Myers, Ph.D.
This reference may be of your interest:
P. Walley (1991) Statistical reasoning with imprecise probabilities.
Chapman and Hall.
Best regards
Antonio Salmeron
-- -------------------------------------- Antonio SalmerónDpt. Estadística y Matemática Aplicada Universidad de Almería Tlf: +34 950 215 669 Fax: +34 950 215 167 --------------------------------------