
Genetic Algorithms Digest   Monday, May 24, 1993   Volume 7 : Issue 14

 - 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 Algorithms vs Tailored Heuristics (2 messages)
	- Evolutionary Structuring of Artificial Neural Networks
	- Re: GP on C or C++
	- AI and Economics: Evolutionary Models in Economics
	- Request for Information on GA'S to Optimize Neural Networks
	- Genetic Algorithms applied to Power Systems Optimization
	- Request for Reprints

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

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

ECAL-93, 2nd European Conference on A-Life, Brussels (v6n31)    May 24-26, 93
CSCS93, 9th Int Conf on control systems & CS, Romania (v7n3)    May 24-27, 93
ANN93, IEE Intl Conf on Artificial Neural Nets, Brighton        May 25-27, 93
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: akonstam@shrew.cs.trinity.edu (Aaron Konstam)
Date: Fri, 14 May 1993 11:29:15 -0500 (CDT)
Subject: Genetic Algorithms vs tailored Heuristics (a response)

It has been 2 months since the submission by James Ignizio in V7 Issue
6 of the GA Digest where he admonished the GA community thusly:
"1. Before proceeding directly to the use of a genetic algorithm
consider doing a simplae literature review so as to establish whether
or not there exists an efficient tailored heuristic approach for the
problem at hand. ....

Believe it or not, I am an advocate of the approach (GA's that is) if
used by the right people, on the right problem, in the right manner."

To me there was an obvious response to these statements. So obvious I
assumed someone would state it. Since they didn't let me state the
response that ocurred to me.

I for one would like to able to know a priori when I am using the
right approach on the right problem in the right manner. If anyone can
suggest how one does that for problems in general I would like to hear
it. I leave to someone else to decide if I am one of the right people.

James Ignizio waited until someone came up with a GA solution and then
found another solution was better. How does one know that the GA
solution is worse before one tries it? 

But more important we the GA researchers (I think anyway) are trying
hard to find a general set of rules to discover in advance whether a
problem will be tractable or intractable to efficient solution to
GA's. No such complete and general theory is available. Until it is we
are forced to try things and see what happens.

If someone knows a better way to solve the combinatorial problem of
the generation of Steiner Triples than the GA based method we have
worked on (Hartley and Konstam, Proceedings of CSC-93, p.366-371) I
would be interested in hearing about it. 

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

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

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

From: "James P. Ignizio" <jpi6g@fulton.seas.virginia.edu>
Date: Sat, 15 May 93 15:39:24 -0400
Subject: GAs vs Tailored Heuristics

GENETIC ALGORITHMS VS TAILORED HEURISTICS --- PART II

James P. Ignizio
University of Virginia
ignizio@virginia.edu

My note on genetic algorithms versus tailored heuristics of about
2 months ago seems to have struck a chord with many readers of
the GA-Digest. Since its appearance, I have received about 50
email messages, with the vast majority of those responding
sharing both my concerns as well as my experiences (i.e., that
tailored heuristics would APPEAR to be superior to GAs when
applied to the specific type of combinatorial optimization
problem for which the tailored heuristic was constructed).

However, there have been a few responses --- notably the recent
one by Aaron Konstam --- that seem to have considered my note to
be a sort of *attack* on genetic algorithms in general. That was
not my intention; rather, I simply have problems with much of the
GA literature with regard to papers on applications of GAs to
various problems of combinatorial optimization. There, one
typically finds a discussion of how an investigator has adapted
the GA method to the solution of a particular type of problem ---
but rarely are such methods compared to existing tailored
heuristics (i.e., methods that appear in the open literature) ---
nor are such alternative methods often even referenced.

The fact is, however, that genetic algorithms is but a heuristic
method itself, albeit a relatively general approach. As such, it
would seem reasonable (at least to me) to attempt to determine
just how efficient the GA method actually is when compared to
alternative methods for solving the particular problem type under
consideration. And I believe that this is a legitimate concern.

I feel that Konstam has ignored this issue in the matters he
raised. However, he does ask some interesting questions. Konstam
asks, for example, just how someone might know a priori just when
he or she is using the right approach on the right problem in the
right manner. He then states *If anyone can suggest how one does
that for problems in general I would like to hear it.*

While such a question is only indirectly related to the concerns
I expressed in my earlier note, I would like to offer a
suggestion. In my discipline we are taught to:

1. First identify the class of problem faced (e.g., traveling
salesperson problem, flow-shop scheduling problem, etc.).

2. Once the specific class of problem has been identified (and
not until then), we are then advised to match that problem type
to any existing methods that have been previously developed for
its solution. If, through either our education or a review of the
literature, we find an existing approach then it is one to at
least be considered for use in solving the problem at hand. If,
however, there are no existing methods for solving such a problem
(a rare occurrence), then we might consider developing a method
for that problem (e.g., adapting genetic algorithms, or simulated
annealing, or tabu search, etc., to the problem at hand).

3. Further, if we use a heuristic method (and remember that
genetic algorithms is but a heuristic method), we are taught to
attempt to evaluate the performance of that method. Thus, should
there be any alternative heuristic approaches (e.g., a tailored
heuristic), then a comparison between the performance of our
approach and the alternatives is usually in order.

This is no difference between this approach and that used (or
that should be used) in the medical profession. If we have
several different medications, all of which may be applied for
the relief of a given disease, then such medications should be
rigorously compared before drawing any conclusions as to which
medication might be *best.* It is my argument that, at this point
in time, we really do not know just how efficient (or
inefficient) GAs are for problems of combinatorial optimization.
Despite this far too many individuals would appear to select GAs
simply for the sake of using GAs.

As such, I would argue that it is essential that the GA-community
seek to determine if my observations (and, evidently, the
observations of many others) with regard to the performance of
GAs versus tailored heuristics is true in general. As I mentioned
in my previous note, my experience has been limited to about 30
cases --- hardly enough to prove the superiority of one approach
over the other, but certainly enough evidence to make one wonder.
Until a thorough study is performed (independently and
scientifically), no one can really say just how good (or bad)
genetic algorithms (or simulated annealing, tabu search, etc.)
actually is for application to problems of combinatorial
optimization.

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

From: I.SANTIBANEZ-KOREF <isk@lautaro.fb10.tu-berlin.de>
Date: Thu, 13 May 93 09:10:41 +0200
Subject: Evolutionary Structuring of Artificial Neural Networks

         *** DO NOT FORWARD TO ANY OTHER LISTS ***

A postscript copy of the following Technical Report can be obtained 
by anonymous ftp at ftp-bionik.fb10.tu-berlin.de 
(ftp-instructions at the end of the message) :

Evolutionary Structuring of Artificial Neural Networks
H.--M. Voigt, J. Born, I. Santibanez--Koref
Technical University Berlin
Bionics and Evolution Techniques Laboratory
Bio-- and Neuroinformatics Research Group

 The report summarizes research on the structuring of Artificial Neural
Networks by  a stochastic graph generation grammar.The main feature of 
the approach is to carry out the graph generation in view of an
individual development process which is embedded in an evolutionary
framework.We explain this approach by  examples, and evaluate 
its practicability.

Comments and questions are welcome. 

==========
ftp-instructions:

   unix %ftp ftp-bionik.fb10.tu-berlin.de
   Connected to lautaro.fb10.TU-Berlin.DE.
   Name (ftp-bionik.fb10.tu-berlin.de:pqp):anonymous
   331 Guest login ok, send your complete e-mail address as password.
   Password:<type your email address here>
   230 Guest login ok, access restrictions apply.
   ftp> cd pub/papers/Bionik
   250 CWD command successful.
   ftp> bin
   200 Type set to I.
   ftp> get tr-02-93.ps.Z
   200 PORT command successful.
   150 Opening BINARY mode data connection for tr-02-93.ps.Z (157041 bytes).
   226 Transfer complete.
   local: tr-02-93.ps.Z remote: tr-02-93.ps.Z
   157041 bytes received in 0.44 seconds (3.5e+02 Kbytes/s)
   ftp> quit
   221 Goodbye.
   unix % zcat tr-02-93.ps.Z | lpr -P<your local postscript printer>

===============
Ivan Santibanez-Koref
FG: Bionik und Evoluionstechnik
FoG Bio- und Neuroinformatik
Sekr. ACK1
Ackerstrasse 71-76
1000 Berlin 6
GERMANY 
Tel.: +49 - 30 - 314 72 677
Fax.: +49 - 30 - 541 98 72
E-mail: isk@fb10.tu-berlin.dbp.de

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

From: tackett@ipld01.hac.com (Walter Alden Tackett)
Date: Wed, 19 May 93 09:50:43 PDT
Subject: Re: GP on C or C++

> I heard GP environment on C or C++ are available.
> If so, how can I get them. 
> 
> Thank you.
> J. Mizoguchi
>
                 SGPC: Simple Genetic Programming in C
                by Walter Alden Tackett and Aviram Carmi
                        (gpc@ipld01.hac.com)

Available via anonymous ftp to sfi.santafe.edu in the directory
pub/Users/tackett
    ^note the caps...

This is a pretty stable version of the code, but not very good
documentation IMHO.  Most of the people we have given it to have been
able to use it though with little or no handholding.  PLEASE send us 
your comments and recommendations to:
gpc@ipld01.hac.com
Time/Life operators are standing by...

What you need to know about the code: it does the same things that
Rice's (aka Koza's) simple LISP does and is set up to handle multiple
populations as well (e.g., for co-evolution).  It is written in C, and
manipulates raw parse-tree structures which may be read and written
in LISP form for backwards compatibility.

You must provide three modules, PROBLEMsetup.c, PROBLEMfitness.c,
and PROBLEMproto.h, where PROBLEM is some descriptive name of the
problem.  E.G., in the version we ship we include REGRESSIONsetup.c and
REGRESSIONfitness.c, Which do Koza's simple regression problem.  We
also include SIN*.c which performs regression on a sinusoid.  We have
also added a crude X interface for the REGRESSION problem.  As a 3rd
example for your enlightenment we include ADFfitness and ADFsetup.
The ADF (Antimean Detection Filter) problem shows you how to build a
simple 2-class "dendritic" classifier described in Experiment 2 of my
ICGA93 paper "Genetic Programming for Feature Discovery and Image
Discrimination in the sfi account).

PROBLEMsetup contains functions to setup the function table, the
terminals table, and code for the functions in the function table.
PROBLEMproto.h contains prototypes for the user defined functions.
PROBLEMfitness contains functions to evaluate and validate populations
and trees, early termination, and definition of the fitness (training
and test) cases.

You should not need to modify any of the other myriad files.

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

From: econec@vax.ox.ac.uk
Date: Thu, 13 May 1993 15:04:13 +0100
Subject: AI AND ECONOMICS: EVOLUTIONARY MODELS IN ECONOMICS

EVOLUTIONARY MODELS IN ECONOMICS AND THE SOCIAL SCIENCES

My working bibliography on this topic is now available by email. It is NOT a
highly polished or anywhere near complete bibliography, though I am in the
process of refining it into sections and providing one line explanations for
why things are on it. It has also absorbed one or two other postings without
acknowledgements, though these will be forthcoming shortly. It may be of
interest to anyone studying CULTURAL EVOLUTION, EVOLUTIONARY MODELS, SOCIAL
DARWINISM (?), GENETIC ALGORITHMS, ALIFE (?), ADAPTIVE SYSTEMS (?), BEHAVIOURAL
THEORY OF THE FIRM, POLITICAL SCIENCE and AI APPLICATIONS generally. My 
interest is in AI techniques and evolutionary applications that actually 
investigate the process of economic action so I am leaving out EXPERT SYSTEMS 
and NEURAL NET PREDICTION which often attempt to predict and prescribe without
explaining, though obviously there are exceptions. (I have tried to provide a
few pointers to this literature however.)

Obviously I will be extremely happy to receive information on further work, to
correspond with interested parties and maybe to provide hints on how to get
hold of some of the stuff, though I shall also be adding contact addresses to
the bibliography in the course of time. I am planning to set up an 
"unofficial" mailing list of interested parties. If there are enough I will try
to get ftp access set up.

Many thanks to those who responded to my original request so long ago!

Edmund Chattoe

Lady Margaret Hall
Oxford
OXON
OX2 6QA

PS If any individuals that I forwarded this to are unhappy about it, my
apologies and please let me know. I was just working from the list of people I
corresponded with who I though wouldn't mind. (Similarly if you are getting
multiple posts, please let me know which address you would prefer me to use!)

PPS Please forward this to anyone you think may not see it and want to. I would
also be happy to hear about any suitable digests or mailinglists who may be
interested.

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

From: Richard J Pryor <rjpryor@somnet.sandia.gov>
Date: Tue, 18 May 1993 08:32:00 -0600
Subject: Request For Information on GA'S to Optimize Neural Networks

Can anyone suggest an introductory paper on the use of
genetic algorithms to optimize the structure of feed-forward
neural networks?
                             
      _/_/_/_/   _/      _/   _/         Richard J. Pryor
     _/         _/_/    _/   _/         Email:  rjpryor@sandia.gov
    _/_/_/_/   _/  _/  _/   _/         (505) 844-2332  FAX  844-2067
         _/   _/    _/_/   _/         Department Manager
  _/_/_/_/   _/      _/   _/_/_/_/   User Support Department (1956)
    Sandia National Laboratories    Albuquerque, NM 87185     

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

From: jvr@micas.inescn.pt (Joao Vasco Ranito)
Date: Mon, 17 May 93 12:18:06 +0100
Subject: Genetic Algorithms applied to Power Systems Optimization

Hello, everybody!

I am looking for some information on Genetic Algorithms applied to
Power Systems Optimization. Does anybody know anything about this stuff?
I was asked by a friend of mine because he is very interested in the
possibility of using GA on this problem. Would you be so kind to
give me some clues?

Thanks in advance,

Joao Ranito

P.S. Feel free to send something by e-mail, instead of bothering
everyone with this particular problem...

Joao Ranito
jvr@micas.inescn.pt
INESC
Porto, Portugal

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

From: intelligent_control@pi-bucuresti.th-darmstadt.de
Date:    Thu May  6 12:59:09 MES 1993
Subject: REQUEST FOR REPRINTS

	My name is Catalin Buiu, and I am Assistant with
the Department of Intelligent Control and Bioengineering
Polytechnical University of Bucharest,Romania.
	My current research interests include Genetic 
Algorithms and Evolution Programming:

A.Theoretical foundations:
	-the study  of convergence
	-possible connections with other nature  inspired
techniques, such as Simulated Annealing
	-messy genetic  Algorithms
	-designing self-adaptive genetic  operators
	-parallel Genetic Algorithms

B.Applications:
	-applications in process identification and control,
especially Genetic Algorithms based fuzzy control  of com-
plex, ill-defined processes, such as bioprocesses
	-applications to neural network problems
	-applications to numerical optimization problems

	In this connection, I'd very grateful in receiving
reprints of papers in the above domains,by regular mail.
	I'd be  interested also in receiving any kind of
software package in the area of Genetic Algortihms.As I can't
receive large files by email, I am ready to send floppies for
receiving the software.

	Thank you very much for your attention to my request.

My mailing address is:
Catalin Buiu
Str.Cetatea de Balta 131
Bl.1 Sc.A Apt.5
77606 Bucuresti 16
Romania
intelligent_control@pi-bucuresti.th-darmstadt.de

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

