
Genetic Algorithms Digest   Monday, 24 September 1990   Volume 4 : Issue 22

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

Today's Topics:

	- Crossing lions with antelopes
	- Evolver by Axcelis, Inc?
	- Questions on Classifier Systems
	- GA's and Discrete Optimization
	- GA Bibliography?
	- ISO...

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

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

Conference on Simulation of Adaptive Behavior, Paris (v4n17)  Sep 24-28, 1990
Workshop Parallel Prob Solving from Nature, W Germany (v4n18) Oct 1-3,   1990
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: Wed, 12 Sep 90 14:54:10 AEST
From: tom@runxtsa.runx.oz.au (Thomas Antalffy)
Subject: Crossing lions with antelopes

    My interpretation of the basic idea of genetic recombination is: "take 
    two good solutions to a problem, and recombine them into a third one". 
    Chances are that this third one will be good as well or perhaps even 
    better than its parents.

    Or will it ? Well, if the search space is reasonably smooth and the 
    two parent solutions were halfway up the hill we are attempting to 
    climb: yes.

    On the other hand there won't be much improvement if the search space 
    is erratic. But this of course is a different question. If our search 
    space is highly non-continuous, chances are that we got the 
    representation seriously wrong in the first place.

    The other main reason for our solution failing would be that the two 
    parent solutions were on two different hills. In other words they are 
    two different solutions to our problem, which can not be recombined. 
    Or, should I say, they are two different species in the same niche. 
    One is a lion and the other is an antelope, both adapted to the same 
    environment, but on their own ways.

    So what's the relevance of this all ? I've just reviewed Classifier 
    Systems and they seem to do a lot of crossing lions with antelopes.

    Classifier Systems that have evolved to a reasonable level of 
    performance feature a complex network of classifiers with various 
    roles. Individual classifiers can be activated under diverse 
    circumstances and have altogether different functionality. One might 
    be a synchronic rule encoding that "big things are dangerous", while 
    the other could be a diachronic one saying "things fall downwards". 
    And still, when it comes to recombination time, if both were useful, 
    we will select them as 'parents' for recombination.

    Are we crossing lions with antelopes in Classifier Systems ?

    Would it not be more useful to find classifiers with similar 
    functionality and recombine them amongst each other ?

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

Date: Tue, 4 Sep 90 13:34:22 EDT
From: Rick_Riolo@um.cc.umich.edu
Subject: Evolver by Axcelis, Inc?

    Does anyone know anything about a commercial package called
    Evolver, by Axcelis Inc?  There was a brief article about
    it in the New York Times 29 Aug 1990.  The article starts:
	 A Seattle company has developed a personal computer
      program that "evolves" the best solution amoung
      various models run on spreadsheets by financial 
      planners for businesses. ... It works as an addition
      to a computer spreadsheet and finds the best [!] 
      answer by using principles taken from theories of 
      natural selection and evolution.
	 These types of programs are known as genetic
      algorithms, and some computer scientists think they
      will be able to use the software to solve a wide
      variety of complex problems beyond the capability
      of traditional mathematical techniques.
    No mention of any computer scientists by name, or anyone else 
    for that matter.  Anyone tried this, or know who is behind it?
      - Rick Riolo

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

Date: Tue, 18 Sep 90 19:23:30 HOE
From: Perfecto Herrera-Boyer <D4PBPHB2%EB0UB011.BITNET@CUNYVM.CUNY.EDU>
Subject: Questions on Classifier Systems

    DEAR COLLEAGUES:

    I AM A PH.D. STUDENT CURRENTLY TRYING TO IMPLEMENT HOLLAND ET AL.'S
    'FRAMEWORK FOR INDUCTION' IN A SIMULATION ENVIRONMENT BASED ON
    CLASSIFIER SYSTEMS.
    THE PROGRAM IS BOTH INTENDED FOR LEARNING ABOUT CLASSIFIER
    SYSTEMS, AND (MAINLY) FOR LEARNING ABOUT COGNITION. IT IS SUPPOSED
    TO BE A PSYCHOLOGICALLY PLAUSIBLE & USER-FRIENDLY COMPUTER ENVIRONMENT
    (THE SYSTEM WILL TRY TO CONJUGATE IDEAS FROM HOLLAND & REITMAN (1978),
    HOHONITH (1986), HOLYOAK, KOH & NISBETT (1989), BILLMAN & HEIT (1988),
    ETC., ).

    I HAVE SOME QUESTIONS THAT I AM NOT ABLE TO SOLVE BY MYSELF, SO I HAVE
    MADE UP MY MIND TO ASK THEM TO YOU. I HAVE TRIED TO SEND THIS QUESTIONS
    DIRECTLY TO JOHN HOLLAND BUT I DON'T HAVE HIS RIGTH E-ADRESS.

    THE SUPPORT OF A CLASSIFIER CX IS DEFINED AS THE SUM OF THE BIDS MADE
    BY THE CLASSIFIERS SENDING MESSAGES THAT SATISFY CONDITIONS OF CX.

    LET {C*} = CLASSIFIERS SENDING MESSAGES THAT SATISFY CONDITIONS OF CX
	       AT TIME T;

    SUPPORT(CX, T) = SUM OF BID(C, T-1) FOR EACH C IN C*


    Question 1:
      WHEN THE MESSAGE ACTIVATING CX IS AN ENVIRONMENTAL MESSAGE (I.E. A
      MESSAGE THAT HAS NOT BEEN POSTED BY A CLASSIFIER IN T-1) WHAT IS THE
      SUPPORT VALUE TO BE ASSIGNED TO CX? IT COULD NOT BE 0 BECAUSE THEN,
      WHEN COMPUTING THE BID IT WOULD BE ALSO 0
      (BID = K x SPECIFICITY x STRENGTH x SUPPORT)
      BUT IT COULD EITHER NOT BE 1, BECAUSE THEN, ENVIRONMENTAL MESSAGES
      COULD BE FAVOURED AGAINST INTERNAL MESSAGES (AT LEAST IF BID IS
      SUPPOSED TO BE BETWEEN 0 AND 1: PERHAPS IT SHOULD EVER BE GREATER
      THAN 1?).

    Question 2:
      IF MY SYSTEM POSTS ONLY 1 MESSAGE AFTER COMPETITION,
      THEN ON SUBSEQUENT 'AUCTIONS' {C*} WILL ONLY HAVE 1 ELEMENT AT MOST.
      THIS DOES NOT SOUND ME WELL IN ORDER TO EXPLOIT THE CONVERGING
      EVIDENCES OF DIFFERENT CLUES PRESENT IN THE ENVIRONMENT...
      WHAT DO YOU THINK OF COMPUTING SUPPORT NOT ONLY FROM THE ACTIVATED
      CLASSIFIERS ON T-1 BUT ALSO FROM THE ALTERNATIVE CLASSIFIERS THAT
      COULD HAVE BEEN ACTIVE IF THEY HAD WON THE COMPETITION?
      EXAMPLE:
	LET US SUPPOSE THAT M0 IS A MESSAGE MATCHING CLASSIFIERS C1, C2,
	& C3.
	  C1 COULD POST MESSAGE M1, C2 COULD POST M2, AND C3 COULD POST M3.
	    M1 COULD MATCH CLASSIFIERS C4 & C6.
	    M2 COULD MATCH CLASSIFIERS C4, C5, & C6.
	    M3 COULD MATCH CLASSIFIERS C5, C6 & C7.
	WHEN COMPETING C1, C2 & C3, THEIR RESPECTIVE BIDS ARE B1 < B2 < B3.
	THE WINNER IS C3, THEN IT POSTS M3 MATCHING C5, C6, & C7.
	NOW C5, C6 & C7 ARE COMPETING:
	- SUPPORT FOR C5 COMES FROM C3, BUT ALSO FROM C2 (BECAUSE IF C2
	  HAD BEEN THE WINNER IT WOULD HAVE POSTED A MESSAGE MATCHING C5);
	- SUPPORT FOR C6 COMES FROM C3, BUT ALSO FROM C1 AND FROM C2;
	- SUPPORT FOR C7 COMES ONLY FROM C3;
	THUS, THE MOST FAVOURED BY SUPPORT WOULD BE C6 (IT RECEIVES MORE
	CONVERGING EVIDENCE).
      WHAT IS WRONG HERE?

    Question 3:
      IF MY SYSTEM COULD POST MORE THAN 1 MESSAGE EACH TIME, WHAT DO YOU
      THINK TO BE THE WAY TO IMPLEMENT IT?
      A) SELECT THE N MESSAGES WITH A HIGHER BID.
      B) SELECT THE N MESSAGES PROBABILISTICALLY
	 (BID WOULD REPRESENT A PROBABILITY OF BEING CHOSEN).
      C) SELECT N MESSAGES, ONE FOR EACH 'CHANNEL' AVAILABLE
	 (I.E. SUPPOSE WE HAVE 3 CHANNELS: A MESSAGE LIST, AN ARM-MOVER,
	 AND A SET OF WHEELS THAT CONTROL ROTATION, THEN WE WOULD SELECT
	 ONE TO BE POSTED TO THE MESSAGE LIST, ANOTHER TO BE POSTED TO THE
	 ARM CONTROLER, AND FINALLY, ONE TO BE POSTED TO THE WHEELS
	 CONTROLER).
      D) ???.


    WELL, PERHAPS IT HAS BEEN ENOUGH FOR TODAY. FORWARD THANKS FOR YOUR
    COMMENTS.

    Perfecto Herrera-Boyer
    <d4pbphb2@eb0ub011.BITNET>
    Dept. Psicologia Basica
    Univ. of Barcelona

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

Date: Sun, 16 Sep 90 14:02:10 EST
From: cpn@cs.purdue.edu
Subject: GA's and Discrete Optimization

    Hi,

    I am interesting for the application of GA's to discrete optimization
    problems (DOP). 

    I would appreciate if you could  send me  a list of : 

      1. conferences
      2. Journals or
      3. other 

    which will help me to find more information on this direction (GA's & DOP).

    Thank you very much

    Nikos Chrisochoides
    Dept. of Computer Science
    Purdue University
    W. Lafayette IN 47906
    e-mail : cpn@sapfo.cs.purdue.edu
    tel    : (317) 4947840

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

From: Mark Hughes <mrh@camcon.co.uk>
Date: Wed, 5 Sep 90 10:14:07 BST
Subject: GA Bibliography?

   Does anyone have a comprehensive machine readable Genetic Algorithm
   Bibliography that they are able to share? I'm thinking of something like
   that in Goldberg's book.

   I would be particularly interested in anything which attempted to classify
   papers by content, as no doubt would many others.

   If such a thing does not exist, then perhaps we ought to be assembling
   one? Any volounteers? I should think that collecting the information
   together initially would be the easy bit, but maintaining and
   distributing it could be onerous.


    ----------------  Eml: mrh@camcon.co.uk or mrh@camcon.uucp
   |   Mark Hughes  | Tel: +44 (0) 223 420024   Cambridge Consultants Ltd. 
   |(Compware & CCL)| Fax: +44 (0) 223 423373   The Science Park, Milton Road,
    ----------------  Tlx: 81481 (CCL G)        Cambridge, UK. (Me, an opinion?)

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

Date: Mon, 10 Sep 90 23:21:46 EDT
From: Yingjia Ding <ding@csgrad.cs.vt.edu>
Subject: ISO...

    Hi,

    Is there anyone who knows one of the following people's email address:
       David J. Powell, Siu Shing Tong (GE company) or 
       Michael M. Skolnick (CS Dept. RPI) ?

    If you know, could you send me a response to 
       ding@CSgrad.cs.vt.edu

    Thank you very much in advance.

    -- Yingjia Ding

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