Re: On web: Clustering Belief Functions (Dempster-Shafer Theory)

Johan Schubert (johans@atlas.sto.foa.se)
Tue, 12 Oct 1999 15:53:54 +0200 (MET DST)

In article <37ff937e.1267862561@news.idt.net>, n.o.s.p.a.m.-mvcs@idt.net (MVCS) writes:
|> [additions added to newsgroup list]
|>
|> schubert@sto.foa.se (Johan Schubert) wrote:
|> >My papers on clustering belief functions, etc., are now available on the web
|> >with URL: http://www.foa.se/fusion/ ...
|>
|> Yes. Very interesting. I have dnld them all and studies some of them
|> (except your thesis, which you have so graciously offered to mail to
|> us.... I posted to you from an alternate address at the SIVAM project
|> site).
|>
|> I would appreciate anyone's comments on the topics covered in these
|> papers as well. I have gathered several other papers showing how the
|> same sorts of things can be accomplished via Bayesian belief net,
|> causal model, and influence diagram methods. I need to know what are
|> the upsides and downsides of the differnt approaches.

Often though, problems are not exactly similar. Then, which metod you
choose depends on your available input data. My clustering approach uses
only "negative" repellent information (the conflict in Dempster's rule)
indicating that pieces of information does not belong together. I.e.
pieces of information in a cluster are not put there because the fit in,
but because they do not fit in anywhere else. This is not the usuall
situation and it tends to be the most difficult case.

If you have additional "positive" information on which pieces of
information belongs together you can develop methods with a better
clustering performance, but then you can not make a direct comparison of
the two methods since the problems solved are different. Then it is more
of a situation where you want to fit a method to a problem rather than
picking the best one of several methods that can solve the same problem.

|>
|> Johan, I added additional newsgroup to the list because you had posted
|> similar messages there in the past, and I wish to elicit (hopefully)
|> some comments from this and other groups.
|>
|> One of my first questions is: are your data sets available for testing
|> and comparison?

My test data set in the neural clustering papers is very simple.
Just generate it: It consists of all 2**n - 1 basic probability assignments
(bpa's) with the different subsets of the frame as propositions and a random
basic probability number (bpn) drawn uniformilly from [0,1]. The frame is,
eg. all integers between 1,n.

Then I use -log(1 - p1 * p2) as a weight between m1 and m2 and start
clustering.

|> Abstracts of some of the data sets? Or does anyone
|> have data sets or know where example data sets exist on-line whereby
|> some of these sorts of ideas can be tested, contrasted, and compared?
|>
|> Cheers!!
|> Mind & Vision Computer Systems
|> "Intelligent Processing Systems for the Energy Industry"
|> _________________________________________________________________
|> | Jeffrey L. Baldwin |
|> | Voice: (281) 752-9099 |
|> | Fax/Data: (281) 752-9101 |
|> | email: mvcs@idt.net 73051.1316@compuserve.com |
|> | http://idt.net/~mvcs |
|> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Johan

############################################################################
This message was posted through the fuzzy mailing list.
(1) To subscribe to this mailing list, send a message body of
"SUB FUZZY-MAIL myFirstName mySurname" to listproc@dbai.tuwien.ac.at
(2) To unsubscribe from this mailing list, send a message body of
"UNSUB FUZZY-MAIL" or "UNSUB FUZZY-MAIL yoursubscription@email.address.com"
to listproc@dbai.tuwien.ac.at
(3) To reach the human who maintains the list, send mail to
fuzzy-owner@dbai.tuwien.ac.at
(4) WWW access and other information on Fuzzy Sets and Logic see
http://www.dbai.tuwien.ac.at/ftp/mlowner/fuzzy-mail.info
(5) WWW archive: http://www.dbai.tuwien.ac.at/marchives/fuzzy-mail/index.html