
Genetic Algorithms Digest   Monday, June 7, 1993   Volume 7 : Issue 15

 - 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 (Info in /pub/galist/FTP)

Today's Topics:
	- Genetic Programming Workshop -- Call For Papers
	- RE: Genetic Algorithms vs Tailored Heuristics (2 messages)
	- Paper on Generating Steiner Triples
	- any ideas for this problem?
	- GAs on large problems -- information query
	- Looking for R.J. Williams and D. Zipser

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

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

ICGA-93, Fifth Intl. Conf. on GAs, Urbana-Champaign (v6n29)     Jul 17-22, 93
COLT93, ACM Conf on Computational Learning Theory, UCSC (v6n34) Jul 26-28, 93
Machine Learning & Knowledge Acq. Workshop (IJCAI), France (v7n1)  Aug 29, 93
IEE/IEEE Workshop on Nat Alg in Signal Processing, Essex (v7n5) Nov 15-16, 93
EP94 3rd Ann Conf on Evolutionary Programming, San Diego (v7n7) Feb 24-25, 94
The IEEE Conference on Evolutionary Computation, Orlando(v7n10) Jun 26-30, 94
SAB94 3rd Intl Conf on Sim of Adaptive Behavior, Brighton(v7n11) Aug 8-12, 94
PPSN-94 Parallel Problem Solving from Nature, Israel (v7n9)      Oct 9-14, 94

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

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

From: Kim.Kinnear@East.Sun.COM (Kim Kinnear - Sun BOS Software )
Date: Tue, 1 Jun 93 15:47:07 EDT
Subject: Genetic Programming Workshop -- Call For Papers

This CFP previously circulated in a variety of forums with May 28 as
the deadline for abstracts.  This has been extended to June 11, due to
difficulty with digestion deadlines for some lists.

However, the first batch of abstracts will go to the committee for
consideration by June 4.   Abstracts submitted between June 4 and June
11 will be considered, but the time available for consideration will
decrease as June 11 approaches, and inevitaby so will the likelihood of
acceptance.  If you are interested in submitting, please submit as soon
as possible using email!

                Call for Papers
 
         Genetic Programming Workshop 
 
	     [sometime during ICGA]
  	         July 17-22
           University of Illinois at
               Urbana-Champaign

The Workshop:

A Workshop in  Genetic Programming will be held during the period of
ICGA-93 July 17-22, 1993 at the University of Illinois at
Urbana-Champaign.  This workshop will bring together researchers
currently involved with genetic programming to share current research
results and work in progress.

Research and results on any facet of genetic programming is welcome,
including the theory of genetic programming, applications to any
problem area, and comparisons to other strategies.  The only
requirement is that the paper/work must touch on genetic programming in
some way.

	Genetic programming (as we define it) is perhaps best defined
	by the book written by John Koza of Stanford (and published by
	MIT Press), called Genetic Programming -- On the Programming of
	Computers by Means of Natural Selection.

	Genetic Programming was defined in the genetic programming FAQ
	this way:

		"Genetic Programmins is the extension of the genetic
		[algorithm] model of learning into the space of
		programs.  That is, the objects that constitute the
		population are not fixed-length solutions to the
		problem at hand, they are programs that, when executed,
		"are" the candidate solutions to the problem.  These
		programs are expressed in genetic programming as parse
		trees, rather than as lines of code.  Thus, for
		example, the simple program "a + b * c" would be
		represented as:

				+
			       / \
			      a   *
				 / \
				b   c

		or, to be precise, as suitable data structures linked
		together to achieve this effect."

	Increasingly, genetic operations on tree structured
	representations are coming to be considered genetic
	programming, since the "language" making up the "program" can
	be very problem specfic indeed.

The workshop will consist of brief presentations of the essential
points of as much work as we can fit in the [as yet unknown] time
available.  Presenters for the workshop will be selected from abstracts
submitted in advance (see the schedule below).  Since the exact time
for the workshop has not yet been determined, we don't know just how
many people we will be able to fit, so this may change as we firm up
the time and date.

The specific location and time of the workshop has yet to be
determined.

A Book of GP Papers:

We are also planning to publish a book of research papers on GP.
Approximately 20 papers of average length 15 pages will be included in
the book.  All abstracts submitted for consideration for the workshop
will also be evaluated for work to be included in the book.  GP
researchers unable to attend ICGA are encouraged to submit an abstract
for inclusion in the book.  Papers for inclusion in the book will be due
by the end of July, and the book would be published in the fall of 1993.

While we have some good interest, we do not yet have a firm commitment
from a publisher.  We will be using the abstracts that we select to
firm up a commitment with a publisher.

The current plan is that papers accepted for the book will be no longer
than 20 pages, (and some had better be shorter), single column, 10
point type.  The final format for the papers accepted for the book is
essentially the same as that used for the Foundations of Genetic
Algorithms 2, by L. Darrell Whitley, just out by Morgan-Kauffman, San
Mateo CA. (1-800-745-7323 to order).  For those of you who just
finished papers for ICGA, the format is similar except that it is
single column.

One of the nice things is that you won't have to include all of that
basic GP explanation boiler plate for *this* book.  More information
will be forthcoming about paper formats, but the plan at this point is
to have every paper in the book include the table that JK uses in "The
Book" to describe each problem, as a way of trying to tie together the
papers in the book into a more coherent whole.

What to submit:

		A one page abstract *as soon as possible*, but not
		after June 11, 1993.

To be considered for the workshop and the book (or either one
separately), you must submit a one page abstract of your work/paper by
as soon as possible, but not after June 11, 1993:

        Kim Kinnear
        Sun Microsystems
        2 Elizabeth Drive
        Chelsmford MA, 01824
        kim.kinnear@sun.com

        (508) 442-0318

Electronic mail is the preferred submission approach, although US mail
is alright.  Postscript over email is ok too, but ASCII is less risky.
Be sure and include all of your contact information, and an email
address if you have it.  Unless you specify, your abstract will be
considered for both the book and the workshop.

If you have a draft of a paper, feel free to submit it -- but you *must*
submit an abstract too.  Certainly a draft of a paper would help the
committee review the contents, but don't assume that you need a draft of
a paper to submit something.  Many of us will have only an abstract.
 
Authors unable to attend the workshop are strongly encouraged to submit 
abstracts for consideration for the book.
 
Abstracts/Papers will be evaluated on significance of results, clarity, and
originality.
 
The Schedule:

May 28, 1993	1 page abstracts due!  [Original date]
June 11, 1993	Final date for abstracts (with decreasing time for
		review and likelihood of acceptance after June 11).

June 25, 1993	Notification of abstracts selected for the workshop,
		and tentative notification for selection for the book.
 
July 17-22, 1993   Workshop in Urbana-Champaign (exact time TBD).

Tentative Schedule:

July 30, 1993   Papers for the book received.
Aug 20, 1993    Review of the papers for the book complete.
Sep 17, 1993    Final camera ready copy RECEIVED.
Nov 1993        Book is available
 
Workshop Coordinator/Editor -- Kim Kinnear
  
Advisory Board -- John Koza, James Rice.

Review Committee -- Peter Angeline, Robert Collins, Kim Kinnear,
Craig Reynolds, Walter Tackett.

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

From: tgd@research.CS.ORST.EDU (Tom Dietterich)
Date: Thu, 27 May 93 09:31:30 PDT
Subject: RE: Genetic Algorithms vs Tailored Heuristics

This discussion remainds me of the old "strong vs. weak methods"
spectrum discussed in Allan Newell's wonderful 1968 paper

  Newell, A. 1969.  Heuristic programming: ill-structured problems,
  in {\it Progress in Operations Research,} Arnofsky, J., (ed.), New
  York: Wiley. 363--414.

Newell sketches a taxonomy of all possible algorithms, arranging them
from general-to-specific.  General purpose algorithms incorporate less
knowledge about the specific task than problem-specific algorithms.
This necessarily makes them less efficient/effective than
problem-specific algorithms.

Nonetheless, there are many reasons to be interested in weaker,
general-purpose algorithms:

  (a) weak methods are easier to apply to a new problem.  Hence, even
  though they may produce an inferior answer, the total time required to
  obtain that answer may be less than the time required to construct a
  more effective, problem-specific algorithm and then execute it.

  (b) weak methods may provide some understanding of biological
  systems.  For example, it is unlikely that biological systems employ
  linear or quadratic programming, but it is much more likely that they
  might incorporate genetic algorithms, simulated annealing, and so on.

  (c) weak methods help us to understand the "space" of all possible
  methods, and this provides us with insight for designing new methods.

So, while I agree wholeheartedly that comparisons between GA's and
other algorithms are important (especially for applications where the
goal is to find the best possible algorithm), it isn't the whole
story. 

It seems to me that a very important problem is to understand how to
incorporate problem-specific knowledge and heuristics into weak
methods such as GA's.  Could a system automatically discover (or
extract) knowledge about a class of tasks that could be incorporated
into its optimization procedures?

Thomas G. Dietterich
Department of Computer Science
Dearborn Hall, 303
Oregon State University
Corvallis, OR 97331-3102
503-737-5559

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

From: SBRUCEID01@cc.curtin.edu.au
Date: 27 May 1993 13:28:02 +0800
Subject: RE: Genetic Algorithms vs Tailored Heuristics

RE: Genetic algorithms vs tailored heuristics  (Doug Bruce)

If you permit, I should like to interpose my own thoughts on this 
subject.

The discussion that I have seen so far emphasizes the practical 
aspects of solving a particular problem as efficiently as possible 
with the available tools. 

But surely we study GA's for other reasons as well; surely many (me 
included) are interested in GA's for their own sake. The GA or 
Evolutionstrategie is a model (grossly simplified, of course) of 
dynamic processes in the external world, so, as such, is worthy of 
considerable attention. A (biological) species which could solve the 
travelling salesperson problem would not have a greatly enhanced 
survival potential and I would not rate a TSP solver as the pinnacle 
of the evolutionary process. But organisms which can formulate the 
problem and want to solve it *were* evolved.

The focus on the efficiency of the GA may lose sight of the far more 
interesting aspects. If there existed a cookbook which we could look 
up when faced with a problem and read off the optimum method to 
apply, I, for one, would lose all interest in that problem. The 
challenge is to apply the GA to novel situations; if it doesn't work 
ask why not?; if it does try to improve its efficiency. If there is 
a better method then by all means apply that to today's industrial 
requirements, but don't forget that today's efficient methods were 
yesterday's inefficient novelties waiting for the correct hardware 
to be built. 

          Doug Bruce,        SBRUCEID01@cc.curtin.edu.au

                             School of Mathematics & Statistics
                             Curtin University
                             Perth, Western Australia

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

From: akonstam@shrew.cs.trinity.edu (Aaron Konstam)
Date: Wed, 2 Jun 1993 13:55:06 -0500 (CDT)
Subject: Paper on Generating Steiner Triples

Due to my mentioning the existance of our Steiner Triples paper we have
received some requests for the paper. So the paper is:
"Using Genetic Algorithms to Generate Steiner Triple Systems",
Procedings of the 21 st. Annual Computer Science Conference, pp.
366-371 (1993).

It can be retrieved as a compressed postscript file called; ga.ps.Z
by anonymous ftp to tusol.cs.trinity.edu in the /pub directory.

Aaron Konstam         
Computer Science
Trinity University
715 Stadium Dr.
San Antonio, TX 78212-7200

telephone: (210)-7367484
email:akonstam@shrew.cs.trinity.edu

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

From: mlevin@husc.harvard.edu
Date: Thu, 20 May 93 18:35:18 -0400
Subject: any ideas for this problem?

     Here's a problem. I have a number of strings. Each string is made
up of a finite (and small) number of symbols. They are of different
lengths, but on average, on the order of 500 symbols or so. I know
that a small subset of these strings belong to a certain group.
However, I do not know why. I conjecture that some pattern within the
string is the deciding factor (for membership in the group or not). I
have some constraints on this pattern (for one, it must be fairly
local, can contain "don't care" symbols, etc.). To prove this, I need
to find the pattern (it need not be 100% successful in delineating
members from non-members, since my original thoughts as to which
strings belong in the group and which don't are based on empirical
data and may also not be 100% correct). 
     What kind of genetic-search, pattern recognition, classifier,
etc. can I use for this? I was thinking of something along the
following lines: make a genetic algorithm system, where the
individuals represent "hypotheses" - each hypothesis is a schema of
symbols which may or may not occur in any given string. The fitness
function is a function of how well this hypothesis tells apart members
from non-members.  Does this seem reasonable? Any advice? Any and all
suggestions, references, etc. will be appreciated. Please email (or
copy) to mlevin@husc8.harvard.edu. 

Mike Levin

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

From: jtoth@obelix.iki.kfki.hu (J. Toth Gabor)
Date: Fri, 28 May 1993 15:21:20 +0200
Subject: GAs on large problems -- information query

Genetic Algorithms seem to be an effective tool to solve many kinds of
optimization problems of decent sizes. However, it seems that GAs, as
many other optimization problems, slow down considerably as the size of
the optimization space grows. In particular, I have seen many examples
dealing with a few ten of real parameters (or a few hundred bits). I
haven't seen, however, attempting considerably larger problems. If you
know of such application I would appreciate if you could send me a
notice either through GA-list or directly to my e-mail address. Negative
answers are also welcome!

Thank you
Gabor J. Toth

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

From: Bernardo Morcego <bernardo@esaii.upc.es>
Date: Tue, 25 May 1993 18:11:45 UTC+0100
Subject: Looking for R.J. Williams and D. Zipser

	Is there anyone out there who knows the email adress of 
Ronald J. Williams and/or David Zipser ?  They used to work in neural
network learning algorithms, and I guess they still do.
Thanks in advance,
					Bernardo.

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


