
Genetic Algorithms Digest   Monday, 5 November 1990   Volume 4 : Issue 26

 - Send submissions to GA-List@AIC.NRL.NAVY.MIL
 - Send administrative requests to GA-List-Request@AIC.NRL.NAVY.MIL

Today's Topics:

	- Updating GA Mailing List (REPEAT)
	- GAs in finance info sought
	- Request for GENESIS downloaded on to a IBM compatible PC
	- Re: alcelis
	- Criteria for evolvability
	- Evolvability and Fitness Noise

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

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

2nd Intl Conf on Tools for AI, Washington, DC (v4n6)          Nov 6-9,   1990
4th Intl. Conference on Genetic Algorithms (v4n17)            Jul 14-17, 1991

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

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

Date: 5 November, 1990
From: Alan C. Schultz (GA-List moderator)
Subject: Updating the GA Mailing List

   We are once again trying to update our mailing list.  The mailing list
   serves two purposes.  One, it allows us to have a land address so that
   conference announcements and other relevant materials can be sent to you.
   Two, it allows us to make sure you have a valid email address, and also
   should the email fail, we can contact you to try to resolve the problem.
   This information will only be used for the above purposes.

   We will be processing your responses electronically, so please follow the
   instructions carefully.

   Copy the lines BETWEEN the two "CUT HERE" lines to a file, and fill in
   each field according to the following list.  Do NOT remove the keys at the
   beginning of the lines.  Leave a space between the key and your response.
   Then mail back the information, with the subject line of your response
   containing "RESPONSE" and nothing else.  Also, do not put ANY other
   material in your mail message, including signature lines.

   IMPORTANT: Mail the response to ga-list-request@aic.nrl.navy.mil.

List of keys and responses:
	%FN 	First name (you could also put middle initial).
	%LN	Last name.
	%AD	Address line. Use as many as you need, adding them if
		neccessary.  Remove those you do not need.
		Make sure to include your country.
	%EM	Email address from the point of view of the internet.
	%PH	Complete phone number.

Example:

%FN Alan C.
%LN Schultz
%AD Code 5510
%AD Naval Research Laboratory
%AD Washington, DC  20375-5000
%AD USA
%EM schultz@aic.nrl.navy.mil
%PH (202) 767-2684

%++++++++++  CUT HERE ++++++++++++++++++++
%FN
%LN
%AD
%AD
%AD
%AD
%AD
%EM
%PH
%++++++++++  CUT HERE ++++++++++++++++++++

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

Date: Tue, 9 Oct 90 07:45 EDT
From: Stephen Downes-Martin <SDMARTIN@rcca.bbn.com>
Subject: GAs in finance info sought

     Any information concerning the use of GAs in the financial domain
     gratefully received.

     Many thanks.

     Stephen Downes-Martin

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

Date: Mon, 15 Oct 90 12:33:22 -0500
From: sam r. thangiah <thangiah@plains.NoDak.edu>
Subject: Request for GENESIS downloaded on to a IBM compatible PC

    Does anybody have the GENESIS system downloaded on to a PC, most
    preferablyon to an IBM compatible PC?  If so, would I be able to
    obtain a copy of it?

    Thanks,

    - Sam

    Sam R. Thangiah,  North Dakota State University.
    300 Minard Hall, Fargo            UUCP:       ...!plains!thangiah
    ND 58105                          BITNET:     thangiah@plains.bitnet
    Office: (701) 237-8199            ARPA,CSNET: thangiah@plains.NoDak.edu

[Editors Note: John Grefenstette's new version of Genesis has been
tested on and is available for PC's.  PLEASE DO NOT CALL JOHN!  His
new Genesis is available as an option with Dave Davis' soon to be
released book on Genetic Algorithms.  I will send out more information
in the next issue of GA-List.]

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

Date: Sun, 21 Oct 1990 14:35:16 PDT
From: John Koza <koza@Sunburn.Stanford.EDU>
Subject: alcelis

    In answer ro Rick Riolo's recent query, the Oct 90 issue of IEEE spectrum
    on page 92 reports on a genetic algorithm spreadsheet called EVOLVER. It
    is written by a Phil Rybeck of Alcelis Inc, 1406 Western Avenue, Seattle,
    Washington 98101, phone 206-624-2446.

    John Koza
    Stanford University

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

Date: Tue, 09 Oct 90 16:34:45 +0100
From: JANSSEN Jacques <CADEPS%BBRNSF11.BITNET@CUNYVM.CUNY.EDU>
Subject: Criteria for evolvability

                   EVOLVABILITY  CRITERIA

    Now  that Artificial Life  and the  GA  are  getting together,  the GA
    community  is  beginning  to  see  more  papers  on the  evolution  of
    artificial  creatures (or BIOTS as  I prefer to  call them), using the
    GA. I shall  give a concrete example  of this  kind of thing  and then
    raise the type of problem I'm increasingly faced with. I use the GA to
    evolve  the  weights of neural  nets   (GenNets) such that  the neural
    outputs perform some  dynamic  control process, e.g.   controlling the
    angles of a pair  of stick legs such that  the legs  walk, or  the leg
    angles of a quadraped such that it can walk, turn, peck, mate etc. The
    GA is used as a means to find adequate  rather than optimal solutions.
    Any reasonable solution is OK, given that the dynamics are too complex
    to analyse. My problem is that sometimes having chosen  my system that
    the  GA  is  to   be  applied  to,   I dont always   get  a sufficient
    evolvability, i.e. the evolution doesnt really get off the  ground. My
    question boils down to what are the criteria  for evolvability so that
    one can choose a  system  appropriately which  evolves.  At the recent
    Paris "Animat"  (Biot)   workshop,  someone  suggested  that  one such
    criterion should be  continuous phenotypic changes for small genotypic
    changes. Ken de   Jong disputes this.    I'm at  present   evolving an
    artificial  nervous system  for a   lizard-like creature called LIZZY.
    Each motion behaviour is controlled   by a  separate GenNet, e.g.  for
    turning left.  The fitness is  the  body  angle rotated  over a  given
    number of cycles.  The neural outputs control  the  angles of the legs
    and one leg moves at a time with 3D rotations  of  the whole so that 3
    legs remain on the floor.  This is the second motion definition I have
    tried because the first, (which  moved all 4 legs simultaneously) gave
    such poor evolvability that I  was forced  to abandon  it. What I need
    are criteria for  good  evolvability to help  me  choose appropriately
    evolvable systems  (or    in my case,   motion  definitions). The only
    criterion I have  heard so far is  the  small genetotypic changes  -->
    small phenotypic changes. This is not true  for LIZZY,  because a tiny
    change in one of the weights in a GenNet can, after many cycles, cause
    the center of gravity of  LIZZY's body to move  a  tiny  bit, but just
    enough to cause the body to tip over onto  another leg.  Hence a minor
    genetic change causes a major  phenotypic (behavioural)  change.    Is
    this important?   Will this  cause  my  fitness landscape  to  be very
    spiky?  My impression is that traditionally, the GA  has  been largely
    used to optimise a given system. What the "GA-Lifers" are trying to do
    is   to  choose systems  which  evolve   well  to  create  the desired
    functionality.  This is an important change of emphasis in GAs. At the
    moment, choosing evolvable systems is very much a hit and miss affair.
    What we need  are  more solid   criteria for  evolvability. Any  ideas
    GAers?

    Hugo de Garis,
     University of Brussels, Belgium  &
     George Mason University, Virginia.

    email:  CADEPS at BBRNSF11.BITNET

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

Date: Thu, 18 Oct 90 19:41:01 +0100
From: JANSSEN Jacques <CADEPS%BBRNSF11.BITNET@CUNYVM.CUNY.EDU>
Subject: Evolvability and Fitness Noise

               EVOLVABILITY AND FITNESS NOISE

    I've been trying to evolve behaviours  in biots (artificial creatures)
    getting  the GA to evolve the  weights of  neural  nets (GenNets) such
    that the neural  outputs generate behaviours (e.g. getting  a biot  to
    walk and turn).  I was wondering why I got good evolvability (i.e. the
    fitnesses grew with the  generations) with one  definition of  fitness
    but not  with  another  (even  though only   a few lines  of code were
    different). It turned out that the second fitness definition showed up
    some subtle bugs  in the program  that were undetected  before.  These
    bugs  meant that the reported fitness  values were noisy, and probably
    too  noisy for  the system to be evolvable.    This may be untrue, but
    nevertheless it raised what I thought might be an interesting question
    for GAers, namely  how  do noisy  fitnesses  affect  evolvability? For
    example, imagine Brooks (MIT Robot Lab) uses a GA technique  to evolve
    behaviours in his Attila  robot (6  two-part legs). Real world fitness
    measurements (distance covered by the robot when evolving  walking, or
    angle rotated when evolving  turning) will not  be precise. There will
    be  errors in the  reported fitnesses. Since  the elite  chromosome of
    each  generation   may differ only minutely   from   the previous best
    (assuming an elitist approach) the new best may be actually worse. How
    bad does the noise have to get before evolvability is killed? How does
    fitness noise influence  evolvability?  Any thoughts?   Are  there any
    references on this  topic?  If the  GA  starts  being  used in Applied
    Artificial   Life and  Robotics,   this  question will  no  longer  be
    academic, because mechanical and sensor fitness measurement is noisy.

      Cheers, Hugo de Garis, Univ of Brussels, Belgium, Europe.
                             George Mason Univ, Virginia, USA.

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