Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!unido!uklirb!shell
From: mgv@usceast.UUCP (Marco Valtorta)
Newsgroups: comp.ai.shells
Subject: Re: Dempster-Shafer theory
Message-ID: <6446@uklirb.informatik.uni-kl.de>
Date: 17 Aug 90 21:49:53 GMT
References: <6390@uklirb.informatik.uni-kl.de> <6441@uklirb.informatik.uni-kl.de>
Sender: shell@uklirb.informatik.uni-kl.de
Organization: University of South Carolina, Columbia
Lines: 65
Approved: shell@informatik.uni-kl.de
Posted-Date: Mon Aug 20 08:22:56 GMT 1990

In article <6441@uklirb.informatik.uni-kl.de> ahlenius@motcid.uucp (Mark Ahlenius) writes:
>sapong@dvinci.usask.ca (Kofi Sapong) writes:
>
>>If anyone has any information on expert system shells that use the
>>Dempster-Shafer approach to handle uncertainty, could he/she kindly
>>pass it on to me. On the other hand if you have or know of any program
>>that implements any form of a propagation scheme using the above 
>>approach, I will interested in having it.
>
>Maybe someone else can comment on this - but I thought that the Depster-
>Schafer theory was considered dead, and that all work was now being
>done in fuzzy and Bayesian type theories.  Please correct me if wrong,
>this is my impression from talking with others at the AAAI-90.
>-- 

IMHO, you are wrong.

There were papers on both the Bayesian and Dempster-Shafer approaches at
AAAI-90 and at the Uncertainty in AI (UAI) conference that took place just
before AAAI-90 at MIT.  (For example, there are at least two articles
on the Dempster-Shafer approach in the AAAI-90 proceedings, by
Halpern and  Fagin (pp.112-119) and by Hsia (pp. 120-125).)

In fact, I was somewhat surprised to see that Dempster-Shafer seems to 
be alive and well.  I attended the closing hours of UAI and witnessed
an argument between Judea Pearl and Philippe Smets.  I will try to give
an impressionistic sketch of what happened.  Smets had presented
a paper on one of the examples presented by Pearl in his book, supposedly
to show the inadequacy of Dempster-Shafer.  Pearl claimed that the proposed
solution required too much designer intervention.  Smets counterargued by
working out his reply in semi-real time.  Pearl listed some pitfalls
of non-monotonic logics that, he claimed, were also pitfalls of 
Dempster-Shafer.  Smets replied that there was no evidence that those
pitfalls could not be avoided in Dempster-Shafer.  He also counterattacked
when, in reply to another participant, he pointed out how fragile the
identification of implication and conditional probability is.

Smets also described an ESPRIT basic research action on uncertain
reasoning called DRUMS ("Defeasible Reasoning and Uncertainty Management
Systems"), which includes research on non-monotonic logics, possibility
theory, and (Dempster-Shafer) belief functions, but in which Bayesian
belief networks have no part.  Since ESPRIT funded a 45 person-years
project that included the development of a large Bayesian network
(MUNIN), it seems that at least one major funding organization has
not decided to abandon Dempster-Shafer.  (Incidentally, a spin-off of
the MUNIN project is the shell HUGIN, the best shell for belief network
building on the market today.)

I hope others will follow up on this topic!
Maybe someone at the funding agencies?  Is Maria Zemankowa at NSF
tuned into this newsgroup?


>===============	regards   'mark  =====================================
>Mark Ahlenius 		  voice:(708)-632-5346  email: uunet!motcid!ahleniusm
>Motorola Inc.		  fax:  (708)-632-2413
>Arlington, Hts. IL, USA	 60004


Marco Valtorta			usenet: ...!ncrcae!usceast!mgv
Department of Computer Science	internet: mgv@cs.scarolina.edu
University of South Carolina	tel.: (1)(803)777-4641
Columbia, SC 29208		tlx: 805038 USC
U.S.A.				fax: (1)(803)777-3065
usenet from Europe: ...!mcvax!uunet!ncrlnk!ncrcae!usceast!mgv