Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!godot.cc.duq.edu!newsgate.duke.edu!news.mathworks.com!newsfeed.internetmci.com!in1.uu.net!01-newsfeed.univie.ac.at!cmir.arnes.si!news.fagg.uni-lj.si!not-for-mail
From: Ales Krajnc <akrajnc@fagg.uni-lj.si>
Subject: New version of aiNet is available
Message-ID: <31E60C31.64F0@fagg.uni-lj.si>
Date: Fri, 12 Jul 1996 10:26:25 +0200
Organization: aiNet
X-Mailer: Mozilla 2.01Gold (Win95; I)
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
Lines: 289

Dear AI community,

I would like to present you a new version of the aiNet
application and encourage you to try it and to tell us your remarks.
It can be found in the following major FTP sites:

http://www.simtel.net/pub/simtelnet/win95/neural/ainet120.zip
ftp://ftp.simtel.net/pub/simtelnet/win95/neural/ainet120.zip

http://www.winsite.com/win95/programr/ainet120.zip
ftp://ftp.winsite.com/pub/pc/win95/programr/ainet120.zip

About the aiNet

The aiNet is a powerful (Windows95) neural network-like
application and is designed specifically to facilitate modeling
task in all neural network problems. It runs under Windows95
(preferable), WindowsNT or Windows 3.1 with Win32s extension.
Here are its major features, which makes it interesting:

List of features from the manual.

The major attribute that distinguishes the aiNet from other ANNs
is the analysis speed. Since the aiNet uses an algorithm, which
does not require learning phase, the answers about prediction can
be obtained almost immediately.

There is also only one coefficient (penalty coefficient), which
has a major effect on the results. If we neglect some aspects, we
can claim that knowing the right value for this coefficient
solves the entire problem. One would now probably expect that it
is hard to determine the optimal value of the coefficient. On the
contrary, it is quite easy! All you need to do is to try
different values and finally select the most successful one.
According to our results there is only one optimal value for the
penalty coefficient. Fortunately, the results are not sensitive
to the optimal value. This means that if you slightly change the
penalty coefficient from the optimal value, the results will not
change much (if at all). Or in other words: The optimal curve is
usually shallow at the optimal point.

Another very important feature is the ability to dynamically
change the "knowledge base". This means that you can add some new
data to  neural network (or remove old ones), add some additional
variables (or remove old ones) and still get answers right away
-- there is no time consuming learning phase.

Our experiments show us that noisy data is the aiNets favorite.
If the data is just noisy, the aiNet will give you excellent
results. When you have chaotic data (and you do not know that),
then you can not obtain any solution, still the aiNet will assure
you  that something is wrong with the data.

The aiNet provides you a way to estimate the rate of error in
your prediction. If your problem is smooth, i.e. without noise,
then this error will represent an estimation for the error in the
predicted result. If you have noisy data, then this will
represent an estimation for the noise around the predicted
result. This means that an error estimation behaves locally.

aiNet is very suitable to work with missing values in your data.
In real life problems it is usually very difficult to find a
perfectly assembled knowledge base -- there is always some data
missing. The aiNet handles missing data automatically and you
need not worry about how to represent such data.

The aiNets graphical user interface is very simple to use. It
looks like a spreadsheet application and if you are familiar with
any other spreadsheet, the aiNet would not present a problem for
you. Almost everything is only a mouse click away, menus are
simple and there are also speed buttons.

On-line help is there for you. If you are stuck and do not know
what to do just press F1 and the aiNet will help you find the way
out.

Several charts are also available. They represent the most
natural way to estimate how good your problem data is. They tell
you visually if your problem can be generalized and what kind of
results you can expect.

With the new two parametric chart you can present calculation
results in 3D space. This gives you a total new look on your
problems.

aiNet provides excellent links to other applications,
spreadsheets mostly. This means that results calculated in aiNet
can be easily transfers to spreadsheets and vice versa. All you
have to do is to Copy and Paste.

Along with the aiNet you get also a development kit build around
the aiNet DLL library. This library includes major aiNet
functions needed to calculate prediction. Now you can create your
own applications based on the algorithm which is used in the
aiNet application. Examples are given for C language and Visual
Basic.


SYSTEM REQUIREMENTS
-------------------

The aiNet requires Microsoft Windows 95 or WindowsNT. (It runs
also under Windows 3.1 with Win32s installed, however minor bugs
were reported.) It requires approximately 9MB of disk space and
8MB of RAM in your system.  Also, a VGA, SVGA, or other high
resolution monitor is recommended. A mouse or other pointing
device is also required.

Secondary (mirror) sites where aiNet120.zip file might be
available can be found below:

(please, do not blame me if the site has not been mirrored yet.)

Australia

-ftp.tas.gov.au
-ftp.iniaccess.net.au
-ftp.bhp.com.au

Austria

-ftp.univie.ac.at

Belgium

-ftp.linkline.be
-ftp.tornado.be

Brazil

-iis.com.br
-unicamp.br

Canada

-direct.ca
-crc.doc.ca

Chile

-sunsite.dcc.uchile.cl

China

-ftp.pku.edu.cn

Czech Republic

-ftp.eunet.cz
-pub.vse.cz
-ftp.zcu.cz

Finland

-ftp.funet.fi

France

-ftp.grolier.fr
-ftp.ibp.fr

Germany

-ftp.rz.ruhr-uni-bochum.de
-ftp.tu-chemnitz.de
-ftp.uni-heidelberg.de
-ftp.uni-trier.de
-ftp.uni-paderborn.de

Greece

-ftp.ntua.gr

Hong Kong

-ftp.cs.cuhk.hk
-sunsite.ust.hk

Italy

-cis.utovrm.it
-ftp.unina.it

Japan

-ftp.riken.go.jp
-ftp.iij.ad.jp
-ring.asahi-net.or.jp
-ring.aist.go.jp
-ftp.saitama-u.ac.jp

Korea (South)

-ftp.sogang.ac.kr

Latvia

-ftp.lanet.lv

Mexico

-ftp.gdl.iteso.mx

Netherlands

-ftp.nic.surfnet.nl

Network

-ftp.synapse.net
-ftp.nuri.net

New Zealand (Aotearoa)

-ftp.vuw.ac.nz

Norway

-ftp.bitcon.no

Poland

-ftp.cyf-kr.edu.pl
-ftp.icm.edu.pl
-ftp.man.poznan.pl

Portugal

-ftp.ua.pt
-ftp.ip.pt

Romania

-ftp.sorostm.ro

Slovenia

-ftp.arnes.si

South Africa

-ftp.is.co.za
-ftp.sun.ac.za

Spain

-ftp.rediris.es

Sweden

-ftp.sunet.se

Switzerland

-ftp.switch.ch

Taiwan

-ftp.ncu.edu.tw
-nctuccca.edu.tw

Thailand

-ftp.nectec.or.th

USA

-ftp.orst.edu
-ftp.digital.com
-uiarchive.cso.uiuc.edu
-ftp.bu.edu
-ftp.cyber-naut.com
-oak.oakland.edu
-mirrors.aol.com
-ftp.rge.com
-ftp.ou.edu
-ftp.cdrom.com
-ftp.hkstar.com

United Kingdom

-ftp.mersinet.co.uk
-sunsite.doc.ic.ac.uk
-micros.hensa.ac.uk
-ftp.demon.co.uk


Sincerely,
Ales Krajnc

