🚀 go-pugleaf

Thanks to (in random order):
I.P. Androulakis
Lester Ingber
Stephanie Leichner
Mark Johnson
James Litsios
Rowan Davies

and to all the other people who showed interest in my request. The messages I
received are (edited):

You might be interested in the following :
"Genetic Algorithms and Simulated Annealing : An
Overview", L. Davis, M. Steenstup, Morgan Kaufman 
Publishers Inc., Los Altos, California, 1987.

Further, I was also interested in comparing simulated
annealing with a genetic based search algorithm that
I developed (for continuous problems). If interested the
details can be found in :
"A genetic algorithmic framework for process design
and optimization", I.P. Androulakis, V. Venkatasubramanian,
Computers and Chemical Engineering, vol 15, no 4, pp 217-228,
1991.

In the IEEE literature, especially in the area of neural
networks design, there has been a lot of work dealing
with the comparison of SA to GA's. It has been over two
years since I stop keeping track of this literature so
I might not be of much help.

I would be more that hapy to share your epxerience in this
subject.

I hope this information would ne of some help,
ioannis

androula@ecn.purdu.edu

I think our article qualifies:
%A L. Ingber
%A B. Rosen
%T Genetic algorithms and very fast simulated reannealing: A comparison
%J Mathl. Comput. Modelling
%V
%P (to be published)
%D 1992

This can be obtained electronically, as saga.ps.Z in the archive:

              Getting (P)Reprints via Anonymous Ftp

     Please  note  that  some of my (p)reprints can be downloaded
via anonymous ftp from ftp.umiacs.umd.edu [128.8.120.23]  in  the
pub/ingber  directory.   If  you  have problems with this, let me
know and I will be glad to prepare a uuencoded copy to  email  to
you.   Sorry,  but I cannot take on the task of mailing out hard-
copies.

     Just follow these procedures on your local machine:
local% ftp ftp.umiacs.umd.edu
[local% ftp 128.8.120.23]
Name (ftp.umiacs.umd.edu:yourloginname): anonymous
Password (ftp.umiacs.umd.edu:anonymous): [type in yourloginname]
ftp> cd pub/ingber
ftp> binary
ftp> get README.file
ftp> get file.ps.Z
ftp> quit
local% uncompress file.ps.Z
local% lpr [-P..] file.ps [to your PostScript laserprinter]

     With a uuencoded copy, first save to  mailfile.   Strip  out
lines  before  "begin"  and  after  "end".   Save to mailfile.uu.
Applying
local% uudecode mailfile.uu
will leave file.ps.Z.  Then, proceed as above.

Lester

  * * * * * * * * * * * * * * * * * * * * * * * * * *
  *                                                 *
  *               Prof. Lester Ingber               *
  *           ____________________________          *
  *                                                 *
  *  P.O. Box 857            [10ATT]0-700-L-INGBER  *
  *  McLean, VA 22101        ingber@umiacs.umd.edu  *
  * * * * * * * * * * * * * * * * * * * * * * * * * *

I'd be very interested if you find anything.  I'm currently looking into
both algorithms, as well as tabu search, and I haven't found a good
comparison between any of them.  I seem to keep running into comparisons
of <insert simulated annealing, genetic algorithms or tabu search here> 
with <insert standard method for solving particular problem>.  I do have
a good overview paper of the 3 algorithms, as well as neural networks (as
applied to optimization) and target analysis.  However, there aren't any
computational results.  I'd be happy to give you what I have, if it would
be of any use.

-- Stephanie Leichner
sleichne@hplabs.hp.com

[Actually this message was interesting because mentioned the tabu search method
which I wasn't aware of. Some tabu search references are given at the end]

Read an article in sci.math.num-analysis and
you will have your answer.

It is article <1992Jul27.154747.167326@ns1.cc.lehigh.edu>
and as you can see (from the first nine characters
of its identifier), it is recent.

mjohnson@netcom.com (Mark Johnson)

[The article anticipated some of the conclusions of the article of L. Ingber
and B. Rosen mentioned above. The anticipated conclusions were that VFSR was
about an order of magnitude faster than GA]

Three or so years back when I was working on automatic placement for integrated
circuit layout I looked in to GA.  A read the literature of the time and the
book you mentioned. My impressions where:

In a non-parallel computing environment GA does no perform better than SA.
In a massivaly parallel environment I can imagine some good results but I
have no experience.

Compared to SA, fine tuning a GA is very tricky.  The problem is to control
the amount of diversity in the population. Too much diversity and the 
crossings result in non-viable solutions, too few and the algorithm loses
any genetic properties and acts like a simple search algorithms.

My conclusion at the time (+3 years ago) was:

It was a new fad that did not give good results but allowed people to publish.

Most hard combinatorial problems are not composite problems that work well
with GA. 

GA is great for students. The concept of a population, the genes, the crossings
offer many fun parameters to experience with.


Note that at the same time (I think), there was some talk about the Tabu 
Algorithm which is a search algorithm with a set of illegal moves (that is
updated).  I never applied it but thought that it looked much more promising
than both SA and GA (for the problems that support it).

I threw away all my references on GA stuff, I might be able to find one or
two references on the Tabu algorithm.

James


-- 
James Litsios                   Phone:       +41 1/256 49 50
Integrated Systems Laboratory   Fax:         +41 1/252 09 94
ETH Zurich                      E-Mail:      litsios@iis.ethz.ch
CH-8092 Zurich, Switzerland    

Regarding your post to sci.math.num-analysis -

I would also be very interested in any results on this subject. I currently
have a book 'Simulated annealing and Genetic Algorithms' borrowed from the
library, but it doesn't seem to have any comparititive study in it.
My impression is that GA's have quite limited application, and gain little
over repeated trials of hill climbing algorithms (but require much more
time in general), though I have very little experience with them.

THANKS ........

Rowan Davies               University of WA
rowan@cs.uwa.edu.au        Pure Mathematics/Computer Science

This is what I've got on the Tabu algorithm:
1)
Future paths for integer programming and links to artificial intelligence
Fred Glover
CAAI Report 85-8
Center for Applied Artificial Intelligence
Graduate School of Business
University of Colorado
Boulder, CO 80309-0419

2)
Interactive decision software and computer graphics for architectural
and space planing
F. Glover, C. McMillan, B. Novick
Annals of Operations Research 5(1985/6)557-573

3)
Using tabu search techniques for graph coloring
A. Hertz, D. de Werra
Ecole Polytechnique Federale de Lausanne, Suisse.


My comments:

1) Looks like a good original reference article.
2) Not much on tabu in it.
3) An application of the algorithm.


I would be interested to hear what final conclusion you make out of your
request.

James

James Litsios                 Phone:  +41 1/256 49 50     |:::::::::::::::::::
Integrated Systems Laboratory Fax:    +41 1/252 09 94     |:::::::::::::::::::
ETH Zurich                    E-Mail: litsios@iis.ethz.ch |::::::::.::::::::::
CH-8092 Zurich, Switzerland                               |:::::::::::::::::::

[Tabu search reference]

I won't draw any conclusion here, but all that I got agrees with what I knew
about the comparison between SA and GA. Any further comment on the subject is
welcome.
Thanks again to everybody.

     Enzo

Enzo Liguori - Canon Information Systems Research Australia