BISC Seminar Announcement, Thursday March 4th, 1999, 4-5pm, 310 Soda

Frank Hoffmann (fhoffman@cs.berkeley.edu)
Wed, 3 Mar 1999 02:10:14 +0100 (MET)

************************************************************
B I S C S e m i n a r A n n o u n c e m e n t
************************************************************

Automatic Pattern Recognition with Hierarchical Fuzzy Rulebases


Speaker :
Rainer Holve

Berkeley Initiative in Soft Computing (BISC)
University of California, Berkeley
E-mail: holve@cs.berkeley.edu

Date: Thursday, March 4th, 1999
Time: 4-5pm
Location : 310 Soda Hall

Abstract

Many training methods for pattern recognition suffer in some way from
the problem that the trained classifiers are not practicable when the
underlying feature space is high dimensional. Some of these problems
will be exemplified as a motivation for a new classification method
that is based on hierarchical fuzzy rulebases (or Hierarchical Fuzzy
Associative Memories - HIFAM). After a brief description of the
training algorithm itself, several properties of the HIFAM-method will
be discussed, e.g. the number of generated rules depending on the
dimensionality of the feature space, the computational effort
necessary to classify a new pattern and the algorithms inherent way of
generalizing the class predictions to regions of the feature space
where no training patterns are present. This last point leads to the
question why it is advantageous to use fuzzy partitions of the feature
space (in contrast to crisp partitions) even if the training algorithm
makes no use of fuzzyness (i.e. the rules learned are initially crisp
and then "fuzzyfied" later) and if the response of the classifier has
to be a crisp class label. Some experimental results will help to
discuss this question. If a new pattern classification method is
proposed, some serious comparisons with other effective state of the
art classifiers should take place (even if such comparisons are
actually difficult problems themselves) to justify the proposal beyond
the pure academic interest. Thus, the results of several benchmark
tests computed with the HIFAM-method are presented and compared to a
large number of results gained from other classification methods. The
presentation finishes with some thoughts about possible future
research topics related to the HIFAM-method.

**********************************************************************
Please direct questions with regard to the contents of the talk
and request for papers to the speaker. To unsubscribe send email
to fhoffman@cs.berkeley.edu NOT !! to bisc-group@diva.eecs.berkeley.edu
**********************************************************************

-- 
---------------------------------------------------------------------------
Frank Hoffmann                               UC Berkeley
Computer Science Division                    Department of EECS
Email: fhoffman@cs.berkeley.edu              phone: 1-510-642-8282
URL: http://http.cs.berkeley.edu/~fhoffman   fax:  1-510-642-5775
---------------------------------------------------------------------------

############################################################################ 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