Newsgroups: misc.forsale,misc.forsale.computers.other.misc,comp.software.international,aus.computers,aus.computers.ai,comp.ai.edu,aus.ads.forsale,comp.ai.shells
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!oitnews.harvard.edu!purdue!lerc.nasa.gov!magnus.acs.ohio-state.edu!math.ohio-state.edu!howland.reston.ans.net!Germany.EU.net!news.mira.net.au!melba.bby.com.au!yarrina.connect.com.au!harbinger.cc.monash.edu.au!lugb.latrobe.edu.au!latcs1!micky
From: micky@latcs1.lat.oz.au (Michaela Petzold)
Subject: Software announcement - AKAT
Message-ID: <DGvJCr.28t@latcs1.lat.oz.au>
Organization: Comp Sci, La Trobe Uni, Australia
Date: Sun, 22 Oct 1995 23:46:02 GMT
Lines: 100
Xref: glinda.oz.cs.cmu.edu comp.software.international:2962 comp.ai.edu:2898 comp.ai.shells:2782


SOFTWARE ANNOUNCEMENT : Automated Knowledge Acquisition Tool (AKAT)
	Version 1.2 Copyright AKT Systems

This notice contains
(i)	a brief overview of AKAT
(ii) 	technical details and
(iii)	a executive summary of AKAT.

Further information can be obtained from:

	AKT Systems Pty Ltd
	P.O. Box 452
	Caulfield East
	Victoria 3145
	AUSTRALIA

or by emailing 

	micky@latcs1.oz.au.

=======================================================================
AKAT - brief overview

Industry and commerce commonly have a large number of databases which 
essentially contain samples of previous decisions or situations.  It would be 
extremely useful to be able to extract the underlying strategy or knowledge 
contained within this data. In this way, one could determine which of the 
attributes have a positive or negative impact on some decision or action.  
The Automated Knowledge Acquisition Tool (ie AKAT) puts together many of the 
major techniques for extracting such knowledge.  These techniques are
essentially simulating the human's capability of learning from experience.  
Given examples of previous situations, AKAT learns the knowledge contained the 
examples and then converts the knowledge into a more useful, concise and 
compact form.  For instance, AKAT can produce rules which can be used in a
decision-support system or an expert system.  AKAT can also be used for data 
reduction as it converts the examples into a more compact and meaningful form.  
In essence, AKAT learns knowledge from examples and then determines the more 
important attributes which uniquely define some concept.

==========================================================================

Technical details

This version of AKAT runs on an IBM compatible PC under windows.  A UNIX 
version will be ready within the next couple of months.

Methods implemented:
-	decision trees (ID3, Tau feature selection criterion)
-	AQ (two versions)
-	neural networks (Hebb's, Delta Rule and Back-propagation)
-	a method of rule extraction using neural networks (ie BRAINNE).

========================================================================
			EXECUTIVE SUMMARY 

The Automated Knowledge Acquisition Tool (ie AKAT) is a package which learns 
knowledge from a database of examples and then converts this knowledge into a 
concise, useful and compact form.  In essence, it determines the underlying 
knowledge contained within the database.

Several techniques exist for learning knowledge from examples.  In order to
use such a technique, one needs to look up the appropriate literature and
then code the technique in some fashion.  Frequently, the descriptions found
are not sufficiently detailed to permit one to code the technique.  This is
both time-consuming and frustrating for someone who simply wants to use the
technique.

AKAT contains several of these techniques already implemented in one package;  
an extra bonus is that there are several neural network implementations also
included.  This is quite unique and useful as in many cases only one technique
is made available as computer code or as a executable. Also, the user interface
of AKAT was specifically designed to be simple and and straightforward.  The 
beta testing of AKAT has indeed shown that little explanation is required for 
the efficient use of AKAT.

AKAT is aimed at three distinct groups:

(1)	people in commerce - these people commonly have large databases of 
	previous experience on issues such as lending, etc. 

(2)	people in industry  - here, large collection of records of previous
	decisions, such as repair or maintenance decisions, are available,

(3)	students and researchers - students in computer science, computer
	systems engineering or cognitive science would benefit from both the
	experimental and teaching value that can be conducted with AKAT.  
	Researchers in universities, research institutes and industry could 
	use AKAT for comparisons and further work.

Instances of the successful use of some of the techniques available in AKAT
include engineers who wish to determine rules for diagnosing engine faults and 
bank managers who wish to determine whether a customer should be given a bank 
loan.  

AKAT has a large potential in many areas of the community. Both industry and 
academic users would find AKAT valuable.  It ease of use enhances its 
attractiveness and usability.
======================================================================

