Newsgroups: sci.image.processing,comp.ai.neural-nets,comp.ai.fuzzy,comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel-eecis!news.mathworks.com!news.kei.com!news.texas.net!news.frontiernet.net!NEWS!news.sprintlink.net!news-dc-5.sprintlink.net!news.sprintlink.net!news-dc-9.sprintlink.net!hpg30a.csc.cuhk.hk!news.cuhk.edu.hk!hkpu01.hkp.hk!usenet
From: "Korris Fu-lai Chung (COMP staff)" <cskchung@comp.polyu.edu.hk>
Subject: RA/PostDoc Position
Content-Type: text/plain; charset=big5
Message-ID: <31D22557.2218@comp.polyu.edu.hk>
Content-Disposition: inline; filename="cfusion.txt"
Sender: usenet@hkpu01.polyu.edu.hk (Usenet Account)
Nntp-Posting-Host: cssolar73.comp.polyu.edu.hk
Content-Transfer-Encoding: quoted-printable
Organization: Hong Kong Polytechnic University
Mime-Version: 1.0
Date: Thu, 27 Jun 1996 06:08:23 GMT
X-Mailer: Mozilla 2.0 (X11; I; SunOS 5.4 sun4m)
Lines: 62
Xref: glinda.oz.cs.cmu.edu sci.image.processing:22776 comp.ai.neural-nets:32199 comp.ai.fuzzy:7706 comp.ai.genetic:9068

		Research Assistant Position Openings =

 =

				in  =

 =

		Department of Computing  =

		Hong Kong Polytechnic University =

		Hunghom, Hong Kong  =

 =


The following funded project is now looking for qualified research personne=
l.  Interested readers may browse the web page http://www.comp.polyu.edu.hk=
/New/RApost.html for further details. =

 =

Project Title :
Classifier Fusion : Theory and Applications  =


Position :
RA/PostDoc for one year, approx. HK$15,000p.m. (~US$1,950p.m.)

Description of Project :
In view of the difficulties of building a single super classifier for the a=
pplications involved, the proposed project aims at establishing a new theor=
y for pattern recognition system design, i.e., classifier fusion.  By takin=
g the advantage of the error independence of different classifiers, classif=
ier fusion is expected to be able to combine the available outputs judiciou=
sly and produce a more accurate output.  Two types of fusion methods will b=
e exploited, namely, integral based and dynamic selection based.  The first=
 one attempts to "average" the classifiers' outputs according to their past=
 performances while the second one works on the idea of making selections a=
mong the classifiers such that the result of the best classifier, for that =
particular input pattern, is outputted.  The developed methods will be appl=
ied to off-line handwritten Chinese character recognition and financial tim=
e series prediction and a class of high performance pattern recognition sys=
tems is expected.  =


Requirement =

(1) A good degree in computing science or related subject;  =

(2) Backgrounds in pattern recognition and neural networks =


Dr.Korris F.L. Chung =

Dept. of Computing =

Hong Kong Poly. Uni.  =




