
Genetic Algorithms Digest   Monday, June 24 1991   Volume 5 : Issue 16

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

Today's Topics:

	- ICGA-91 Workshop on GAs and Machine Learning
	- Re: Grey Coding
	- invalid offspring
	- ICGA-91 Workshop on biological influences, more details

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

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

4th Intl. Conference on Genetic Algorithms (v5n9, v5n14)     Jul 13-16, 1991
AAAI 91, National Conference on AI, Anaheim, CA              Jul 14-19, 1991
EUROPEAN SUMMER SCHOOL on MACHINE LEARNING (v5n7)            Jul 22-31, 1991
Genetic Algorithm Course, Stanford (v5n11)                   Jul 22-26, 1991
IJCAI 91, International Joint Conference on AI, Sydney, AU   Aug 25-30, 1991
First European Conference on Artificial Life (v5n8)          Dec 11-13, 1991
ECAI 92, 10th European Conference on AI (v5n13)              Aug  3-7,  1992

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

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

From: gref@AIC.NRL.Navy.Mil
Date: Thu, 20 Jun 91 09:30:45 EDT
Subject: Workshop on GAs and Machine Learning


		      Workshop on GAs and Machine Learning
				Monday, July 15
			       4:15 pm -- 6:00 pm

    I will be hosting a workshop on GAs and machine learning at ICGA-91.
    The workshop will focus on the application of GAs to problems that have
    been addressed by more traditional symbolic machine learning techniques,
    including concept formation from examples, conceptual clustering,
    explanation-based learning, and learning policies for sequential
    decision problems.  Other topics of interest include the applications of
    of PAC analysis methods to GAs, and the relationships between GAs and
    classifier systems and other methods of reinforcement learning (e.g.,
    Q-learning).

    The format will be informal.  If you would like to participate, let me
    know and I will add you to the mailing list.  If you would like to
    discuss your work in this area, please send me a note as soon as
    possible so that I can construct a schedule.

    John Grefenstette
    gref@aic.nrl.navy.mil

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

From: vose@cs.utk.edu
Date: Fri, 14 Jun 91 14:37:32 -0400
Subject: Re: Grey Coding

    In the GA-List v5n15, there was an algorithm for grey coding which
    is quite nice.  However, the impression was left that the matrix
    technique would be inefficient.  This is not true since the arithmetic
    is modulo 2.  Entire words of data can be efficiently processed using
    exclusove-or.

    Infact, if one is careful about exploiting the linearity of matrix
    multiplication to build and take advantage of a precomputed table,
    then any n-bit vector can be multiplied by a fixed matrix by XOR'ing
    together n/8 items from the table.  This can be improved further if
    a table with more than 256 entries is used.

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

From: Leslie Burkholder <lb0q+@andrew.cmu.edu>
Date: Mon, 10 Jun 91 16:46:04 -0400 (EDT)
Subject: invalid offspring

    We have a problem about generating invalid offspring at one stage in
    running a GA. Has anyone suggestions?

    Here are the details.
    Parents are basically finite automata. They represent strategies for
    playing iterated games (eg, prisoner's dilemma, chicken, etc.). When
    successful parents are combined, sometimes invalid offspring are
    produced. The invalid offspring are syntactically incomplete. They
    include states from which there are no transitions.

    We can see two ways of guaranteeing valid offspring:

    (1) To preprocess all the members of the initial parent generation so
    that they will, when combined with one another, generate no invalid
    offspring. One way to do this (the only way?) is to have each member of
    the initial parent generation and every subsequent generation have
    exactly the same number of states. This produces valid offspring. But we
    see no way to do this automatically or mechanically and yet not
    potentially alter the behavior of some members of the parent generation.
    Certainly we can do this by hand. But we suspect that in some cases
    there will be no way to do this, by hand or automatically, without
    altering the behavior of the automaton. Think of an automaton that uses
    its states to count.

    (2) To fix the invalid offspring after they are generated. This can be
    done in various ways: States in the invalid offspring that have no
    transitions are all pointed back to the initial state, or they are all
    pointed back to themselves, or they are pointed to arbitrary other
    states in the automaton. Unfortunately, none of these really seems to be
    a better repair than the other.

    Thanks
    Leslie Burkholder
    Carnegie Mellon University

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

From: schultz@AIC.NRL.Navy.Mil
Date: Mon, 17 Jun 91 22:33:16 EDT
Subject: ICGA91 Workshop on biological influences, more details


    PURPOSE

    The workshop is intended to bring together GA researchers who are
    interested in improving the GA's performance in problem solving with
    mechanisms that are analogical to, or inspired by biological mechanisms.

    Here is a short list of the type of mechanisms we are interested in:

    niche formation and speciation, crowding factors, sharing functions,
    mating restrictions, diploidy and dominance, overlapping generations,
    performance of individual (super-being) vs performance of society,
    competition vs cooperation, context sensitive evaluation (i.e. evaluating
    all individuals in the population together, such that any individual's
    performance is affected by others).


    FORMAT

    The workshop will consist of a presentation session followed by a
    discussion.

    Presentations:

    Presentations will be very short, and should be treated as "verbal
    abstracts."  The purpose is to give the audience a general idea of the
    type of mechanism being used, but to avoid to much detail.  If members of
    the audience like the idea, they can contact the author for more detail.
    This way, we can have more presentations, and therefore be exposed to
    more ideas.  I would like to limit each presentation to 5-15 minutes
    depending on the number of people interested in speaking.

    The presentations will be followed by a general discussion on biologically
    influenced mechanisms: what has proven useful in the past, what is new,
    and what we would like to see down the road.


    RESULTS of workshop

    We are attempting to provide the following at the workshop, or shortly
    thereafter:

    1) A bibliography of related work. To that end, please send references of
    work in this area, either your own or other works that you feel are
    important to include.

    2) Workshop "proceedings" with abstracts or initial papers from work
    discussed at the workshop.

    3) Summary of discussion session.


    PARTICIPATION

    If you are interested in giving a presentation, please send (email) a very
    short description of work to be discussed at the workshop. Can you send an
    abstract or paper on the work before the workshop, or will one be
    available by the workshop?  Also remember to send as many references as
    possible to include in the bibliography.

    If you would just like to attend, please let me know so we can have some
    idea of the number of people we might expect. You can also send references
    for the bibliography!


    Please send the requested information, and any questions you may have to
    me at schultz@aic.nrl.navy.mil  (202) 767-2684.


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