
Genetic Algorithms Digest   Wednesday, July 8 1992   Volume 6 : Issue 23

 - Send submissions to GA-List@AIC.NRL.NAVY.MIL
 - Send administrative requests to GA-List-Request@AIC.NRL.NAVY.MIL
 - anonymous ftp archive: FTP.AIC.NRL.NAVY.MIL (see v6n5 for details)

Today's Topics:
	- Administrivia
	- Goldberg's book and Classifiers
	- Classifier Systems
	- GAs and Australians (2 articles)
	- searching for multiple solutions in multi-modal search spaces
	- The role of mutation????
	- TR: CLSs and behavior-based robotics paper available 
	- TR: GIGA Reports

**********************************************************************

CALENDAR OF GA-RELATED ACTIVITIES: (with GA-List issue reference)

 10th National Conference on AI, San Jose,                    Jul 12-17, 1992
 FOGA-92, Foundations of Genetic Algorithms, Colorado (v5n32) Jul 26-29, 1992
 COG SCI 92, Cognitive Science Conf., Indiana, (v5n39, v6n22) Jul 29-1,  1992
 ECAI 92, 10th European Conference on AI (v5n13)              Aug  3-7,  1992
 Parallel Problem Solving from Nature, Brussels, (v5n29)      Sep 28-30, 1992
 SAB92, From Animals to Animats, Honolulu (v6n6)              Dec  7-11, 1992
 Intl. Conf. on Neural Networks and GAs, Innsbruck (v6n22)    Apr 13-16, 1993

 (Send announcements of other activities to GA-List@aic.nrl.navy.mil)

**********************************************************************
----------------------------------------------------------------------

From: Alan C. Schultz (GA-List Moderator)
Date: Wednesday, July 8 1992
Subject: Administrivia

    1) PRIORITY OF SUBMISSIONS:

    We are experiencing growing pains.  We have gone from 300 to 1300 readers
    in several years.  The number of new researchers in the area of genetic
    algorithms has grown tremendously recently, and this has led to more
    submissions to ga-list.

    Occasionally people want to know why an announcement they sent to ga-list
    has not been published within a certain amount of time.  Also, we are
    starting to get more and more conference announcements and announcements
    of technical reports, some of which are very long or not relevant to
    genetic algorithms.  I want to avoid sending out issues too often or
    sending out issues with information of low interest to the group in
    general.  Therefore, I am going to step out on a limb, and describe my
    priorities for publication of articles and also suggest guidelines
    (policies?) for submissions of conference and technical report
    information.

    Conference, workshop and symposium announcements: First priority among
    conference announcements are those that explicitly mention interest in
    GAs.  Time deadlines are considered.  A very important consideration is
    length of the article.  So as not to clog mail exchangers, we limit the
    size of each issue.  A very long conference submission that is only
    marginally related to GAs will probably not go out on the list quickly
    unless there is a lack of regular articles to publish.  In some cases, it
    may not go out at all if it has been cross-posted, is very long or there
    are lots of relevant articles to get out.  The bottom line on conference
    submissions: keep them short, and give an (e)mail address for people to
    write for more information.  Remember, we maintain an ftp site for
    ga-list.  In the directory /pub/ga-list/information/conferences I keep the
    full text to announcements for conferences, so please feel free to send
    the full text of your announcement to ga-list-request, asking that it be
    put on the ftp site.  Then send a short, to the point version to ga-list
    that will go out on the list.  I will attach a note to the
    message describing where it can be found on the ftp server.

    Technical report announcements: We feel that publishing information on new
    technical reports in an important service to you.  However, in some cases,
    we have received TR announcements that have too much detail on all the
    various ways to get it over the net, etc.  This detail makes it rather
    long, which as noted above, can delay it getting out when compared to
    other articles.  If someone has trouble getting your technical report,
    that DO have your address, and they can contact you directly to get the
    details on other ways to get the article.  Bottom line: keep it short and
    assume people know how to use ftp, etc.

    In all cases, technical discussions will have top priority on the list,
    followed by technical report announcements, followed by conferences.

    2) CONCERNING FTP ACCESS:

    The correct internet address to use for
    ftp.aic.nrl.navy.mil is 192.26.18.74.  The other address USED to be
    correct, but we changed the ftp server.

    --Alan C. Schultz

------------------------------

From: Keith Chalk <kwc@information-systems.east-anglia.ac.uk>
Date: Fri, 3 Jul 92 10:46:04 BST
Subject: Goldberg's book and Classifiers

   I am currently implementing a Classifier System which uses a Genetic
   Algorithm and parallelism on two different platforms (Mac and UNIX). I use
   a bucket brigade type algorithm, and used the example from "Genetic
   algorithms in search, optimisation and Machine Learning", (p227). My
   results differ to those in the book, and although it is only a small
   example, I would like to know if there is a mistake, or if I am doing
   something wrong. The end strengths are given as: 220,208,196,206. I have:
   220,218,196,196, this is without any GA code being used. Is this just a
   typo, or have ten 'weights' been redistributed for some purpose??  Also,
   does anyone have any test data for checking the workings of Classifier
   Systems?  Thanks for any help you can give,

   -Keith.

------------------------------

From: Cathy Escazut <escazut@mimosa.unice.fr>
Date: Wed, 3 Jun 92 15:17:28 +0100
Subject: Classifier Systems

	I'm a PhD student at the University of Nice-Sophia Antipolis
	(France). I am currently working on specialization and
	generalization in classifier systems and I would like to test my
	results by teaching the system how to behave on a circuit.
	If you have done work in this area or even heard about it please
	send me mail. 

	Thanks in advance.


	Cathy.
	escazut@mimosa.unice.fr

------------------------------

From: kcj@matilda.vut.edu.au (Kathleen C Juliff)
Date: Sun, 21 Jun 92 20:27:16 EST
Subject: GAs and Australians

   Sourav and I would like to make contact of people researching in the area
   of genetic algorithms in Australia. We have in mind forming some sort of
   discussion group. We are both actively researching in the area. We think
   it could be well worth while for others like us, in Australia, to have
   some sort of contact.

   Anyone interested could contact Sourav (sourav@archsci.arch.su.OZ.AU) or
   myself, indicating research interests and ideas for an Australian group.

   My area is GAs and optimisation, specifically GA hybrids.  Sourav's is GAs
   and applications to Design.

   Kate and Sourav.

   kcj@matilda.vut.edu.au
   sourav@archsci.arch.su.OZ.AU

[Editor's Note: Due to various delays, this message did not get out for 2
1/2 weeks.  But look at the next message; it looks like they are making
good progress!]

------------------------------

From: ausga@matilda.vut.edu.au (Australian GA group)
Date: Wed, 8 Jul 92 10:38:30 EST
Subject: Re: GAs and Australians

   An Australian group, AUSGA, exists in order to network
   Australian researchers interested in GAs. We have about
   20 members, but feel there may be more out there. If
   any person living in Australia would like to join our
   group, please email

   ausga@matilda.vut.edu.au

------------------------------

From: ELO Sara <elo@cui.unige.ch>
Date: Mon, 22 Jun 1992 15:22:19 +0200
Subject: searching for multiple solutions in multi-modal search spaces

Hello,

   As I am working on my diploma on parallel GAs, I would like to know what
   has been done/published on searching for multiple solutions in highly
   multi-modal search spaces. Apart from the fitness sharing method or
   simply simulating isolated runs, is there existing literature on other
   methods?

   Thank you very much in advance.

   Sara Elo
   elo@cui.unige.ch

------------------------------

From: kcj@matilda.vut.edu.au (Kathleen C Juliff)
Date: Wed, 1 Jul 92 17:29:20 EST
Subject: The role of mutation????

   I have developed a GA for an optimisation problem.  The problem is that of
   arranging layers of cartons on a palletised truck, subject to seven
   contraints.  There are typically between 80 and 100 to be arranged on the
   truck, in 8 to 12 pallets. Layers must be placed on pallets, and pallets
   must be arranged in an order to optimise the load.

   The chromosomes are strings of integers, representing layer number.

   I am using uniform order-based cross-over.

   The GA works well, but I have found that neither scrambled list, nor
   position-based mutation are of any help. In fact the load is marginally
   better when mutation does not occur at all.

   Has anyone else had this happen? Has anyone any ideas as to why it may be
   happening?

------------------------------

From: dorigo@ICSI.Berkeley.EDU (Marco Dorigo)
Date: Sun, 21 Jun 92 16:19:04 PDT
Subject:  CLSs and behavior-based robotics paper available 

   The following technical report (to appear on IEEE Transactions on Systems,
   Man, and Cybernetics, Vol.22, No.6, November 1992) is available and can be
   requested to:

   Marco Dorigo
   International Computer Science Institute
   1947 Center Street
   Suite 600
   Berkeley, California 94704-1105
   e-mail: dorigo@icsi.berkeley.edu

   Title:

   Genetics-based Machine Learning and Behaviour Based Robotics: A New
   Synthesis

   by Marco Dorigo and Uwe Schnepf 

   Abstract

   Intelligent robots should be able to use sensor information to learn how
   to behave in a changing environment. As environmental complexity grows,
   the learning task becomes more and more difficult. We face this problem
   using an architecture based on learning classifier systems and on the
   structural properties of animal behavioural organization, as proposed by
   ethologists. After a description of the learning technique used and of the
   organizational structure proposed, we present experiments that show how
   behaviour acquisition can be achieved. Our simulated robot learns to
   follow a light and to avoid hot dangerous objects. While these two simple
   behavioural patterns are independently learnt, coordination is attained by
   means of a learning coordination mechanism. Again this capacity is
   demonstrated by performing a number of experiments.

------------------------------

From: Joe Culberson <joe@cs.ualberta.ca>
Date: 	Fri, 26 Jun 1992 15:01:47 +0100
Subject: GIGA Reports

  Three technical reports on gene invariant genetic algorithms are
  available.  A subclass of the class of genetic algorithms, this
  algorithm and its variations represent a unique approach with many
  interesting results.  The primary distinguishing feature is that when a
  pair of offspring are created and chosen as worthy of membership in the
  population they replace their parents.  With no mutation this has the
  effect of maintaining the original genetic material over time, although
  it is reorganized.
  
  These papers may be obtained by email to britta@cs.ualberta.ca
  or you may obtain the compressed postscript form via anonymous ftp to
  	thorhild.cs.ualberta.ca
  cd to directory pub and select the reports corresponding to the
  numbers below.
  
  A program implementing the GIGA algorithm may be obtained from the
  directory pub/GIGA. See the README file for details. The program is
  made available to allow the reader to test various strategies and
  results mentioned in technical report TR92-02. The interface is
  somewhat primitive at this point. TR92-06 is a description of the 
  of the program operation.
  
  Comments should be emailed to joe@cs.ualberta.ca
  
  Following are the abstracts of the three papers.
  
  TR 92-05 Genetic Invariance: A New Type of Genetic Algorithm
  (Master's Thesis title: Genetic Invariance: A New Approach to
  Genetic Algorithms)
  
  author: Michael Lewchuk
  April 1992
  
  Genetic algorithms are adaptive search algorithms which generate and
  test a population of individuals, where each individual corresponds to
  a solution.  They then adapt to the information obtained from testing,
  seeking superior solutions by selecting and combining solutions of
  above average value.  As the number of superior individuals in the
  population increases, the number of inferior individuals decreases.
  This thesis introduces Genetic Invariance, a similar family of generate
  and test problem solvers which uses a different selection and
  replacement strategy. In the best case, it achieves superior solutions
  without eliminating inferior characteristics.   Although
  characteristics may initially be associated with inferior solutions,
  they may prove to be superior when combined with other particular
  characteristics.  Mathematical analysis of lower bounds of Genetic
  Invariance on a simple function is given, and several properties of
  Genetic Invariance are explained using this analysis.  A comparison and
  contrast is done to show how the two selection strategies achieve
  optimization in different ways.  An analysis of the assumption and
  strategies of each system explains likely beneficial and detrimental
  effects of each system, while empirical analysis is given which
  demonstrates these effects.  Together, they show each system's features
  and drawbacks.
  
  
  TR 92-06 GIGA Program Description and Operation
  
  author: Joseph Culberson
  June 1992
  
  This document describes the gene invariant genetic algorithm (GIGA)
  program.  This program represents a unique approach to designing GAs
  with many interesting results.  The primary distinguishing feature is
  that when a pair of offspring are created and chosen as worthy of
  membership in the population, they replace their parents.  In the
  absence of mutation, this has the effect of maintaining the original
  genetic material over time, although it is reorganized, and hence the
  ``invariant'' in the name.
  
  The source code for the GIGA program, written in the programming
  language C, is available on line.  This document explains how to use
  this program and describes the inputs.  It also discusses the design
  philosophy and indicates several possibilities for future extensions
  and variations.  
  
  
  TR 92-02 Genetic Invariance: A New Paradigm for Genetic Algorithm Design 
  
  author: Joseph Culberson
  June 1992
  
  This paper presents some experimental results and analyses of the gene
  invariant genetic algorithm(GIGA). Although a subclass of the class of
  genetic algorithms, this algorithm and its variations represent a
  unique approach with many interesting results.  The primary
  distinguishing feature is that when a pair of offspring are created and
  chosen as worthy of membership in the population they replace their
  parents.  With no mutation this has the effect of maintaining the
  original genetic material over time, although it is reorganized.
  
  In this paper no mutation is allowed.  The only genetic operator used
  is crossover.  Several crossover operators are experimented with and
  analyzed.  The notion of a family is introduced and different selection
  methods are analyzed.
  
  Tests using simple functions, the De Jong five function test suite and
  several deceptive functions are reported.  GIGA performs as well as
  traditional GAs, and sometimes better.  The evidence indicates that
  this method makes more effective use of the crossover operator, in part
  because it never loses genetic material and thus has greater scope for
  recombination.
  
  A new view of crossover search space structures and approaches to
  analysis are presented.  Traditional methods of analysis for GAs do not
  seem to apply since GIGAs cannot be said to give increased trials to
  the best schemata in the usual sense.  However, the analysis of
  crossover search space structures may have applications in traditional
  GA analysis.
  -- 
  		Joe Culberson
  		joe@cs.ualberta.ca
  		(403) 492-5401

------------------------------
End of Genetic Algorithms Digest
******************************
