Fuzzy-Clustering

[Arturo]

Hi collegues,

I am working on my PhD in odour recognition. Now I am working with FCM
to cluster aromas. My data is 3-5 dimensional and the number of clusters
is 3-6.
I think the best solution is to change typical point prototypes to use
different ones as lines and circles in an n-dimensional space.
Is there someone who can give me a reference for this problem, articles,
algorithms, software,etc. ?

I would appreciate very much your help.

Regards from Barcelona,

--
Arturo Ortega Mansilla

Electronic Engineering
Dept. Electronica
Universitat de Barcelona
C/Marti i Franques, 1
08028 Barcelona (Spain)

Fax +34 93 4021148
Tlf +34 93 4029070
Email ort...@el.ub.es

############################################################################
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 list...@dbai.tuwien.ac.at
(2) To unsubscribe from this mailing list, send a message body of
"UNSUB FUZZY-MAIL" or "UNSUB FUZZY-MAIL yoursubs...@email.address.com"
to list...@dbai.tuwien.ac.at
(3) To reach the human who maintains the list, send mail to
fuzzy...@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

[jaw...@iname.com]

I don't find any reason for you to go for higher degree prototypes for
your classification. Points (FCM) should be sufficient.

Anyway, there are a number of interesting algorithms using lines and
circles as the prototypes.

1. Fuzzy C Varieties (lines) - J. Bezdek
2. Fuzzy C elliptotypes(lines + points) - J. Bezdek
3. Fuzzy C Shells - R. Krishnapuram /R. N. Dave

In case, you require any further information, mail me at
jaw...@cair.res.in

Jawahar

In article <369E3011...@el.ub.es>,

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own

[Robin B. Lake]

> Hi collegues,
>
> I am working on my PhD in odour recognition. Now I am working with FCM
> to cluster aromas. My data is 3-5 dimensional and the number of clusters
> is 3-6.
> I think the best solution is to change typical point prototypes to use
> different ones as lines and circles in an n-dimensional space.
> Is there someone who can give me a reference for this problem, articles,
> algorithms, software,etc. ?
>
> I would appreciate very much your help.
>
> Regards from Barcelona,
>
>
> --
> Arturo Ortega Mansilla
>
> Electronic Engineering
> Dept. Electronica
> Universitat de Barcelona
> C/Marti i Franques, 1
> 08028 Barcelona (Spain)
>
> Fax +34 93 4021148
> Tlf +34 93 4029070
> Email ort...@el.ub.es
>
>

I suggest you look at "Information Theory and Esthetic Perception"
by Abraham Moles for background on the fundamental sensory problem.

For a primitive start on displaying circles in color with labels, there
is the famous "sillyballs.c" program by Bo3b Johnson, circa 1988. I just
toyed with it as part of the Metrowerks C compiler package.

Cheers,
Rob Lake
Environmental Modeling Inc.
r...@po.cwru.edu