Genetic Algorithms Digest    Friday, 9 Sep 1988    Volume 2 : Issue 20

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

Today's Topics:
	- GAs for Floorplanning
	- Online and Offline Performance (2)
	- 1989 GA Conference Dates Set

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

Date: Tue, 6 Sep 88 15:13:37 edt
From: Kurt Thearling <steinmetz!vdsvax!thearlin@uunet.UU.NET>
Subject: ga's for floorplanning

There is a paper to be published at the 1988 International 
Conference on Computer-Aided Design (ICCAD-88) that some members 
of this mailing list might find interesting.  The conference runs from
Nov. 7-10 in Santa Clara, CA.

The paper is titled "Floorplan Design Using Distributed Genetic 
Algorithms" and is authored by J.P. Cohoon, et.al.  I haven't
seen a preprint yet but if you are interested in obtaining a copy
you can probably get one from Dr. Cohoon (cohoon@virginia.edu).

kurt

-----------------------------------------------------------------------
Kurt Thearling                               kurt%bach@uxc.cso.uiuc.edu
University of Illinois                           kurt@bach.csg.uiuc.edu
Coordinated Science Laboratory      kurt@bach.csg.uiuc.edu@relay.cs.net
1101 W. Springfield Avenue            kurt@bach.csg.uiuc.edu@uunet.uucp
Urbana, Il  61801                     thearlin%vdsvax.tcpip@ge-crd.arpa 
(217) 333-7914                         thearlin@vdsvax.steinmetz.ge.com
-----------------------------------------------------------------------


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

Date: Tue, 30 Aug 88 18:39:10 +0200
From: prlb2!vub.vub.UUCP!arti1!bernard@uunet.UU.NET (Bernard Manderick)
Subject: online and offline performance

One of my collegues and I are experimenting with variants of the genetic
algorithm. We want to compare our results with those reported in the
literature. Hence, we are using the classical notions of online and offline
performance. The papers we have at our disposal and where these notions are
used are not very clear about what is ment by function evaluation. 

New individuals are obtained by selection, crossover and mutation. Which
ones of these operations are taken into account in determining the number of
function evaluations? Is it CROSSOVER and MUTATION 
OR CROSSOVER, MUTATION and SELECTION.

Consider the case of a population of 50 individuals, a crossover rate of 60%
and no mutation. If crossover, mutation and selection determine the number
of function evaluations then we will have after 100 generations 5000
evaluations. If only crossover and mutation determine the number of function
evaluations then we will have after 100 generations about 3000 evaluations. 

From the available literature, we could not decide which one of the above
interpretations is correct. However, the correct interpretation is very
important for comparing results. Can you give us the correct interpretation?

Bernard Manderick.

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

Date: Fri, 2 Sep 88 08:05:30 EDT
From: John Grefenstette <gref@nrl-aic.arpa>
Subject: Re:  online and offline performance

In my genetic algorithms systems, the evaluation of individuals
is a separate step from selection, mutation or crossover.  The
philosophy is to evaluate only the "new" individuals formed during
each generation.  For example, if the population has 50 individuals
and the crossover rate is 60% and mutation rate is 0, then crossover
produces 30 new structures that are marked for evaluation.  The other 20
are clones from the previous generation, and need not be evaluated
again if the evaluation function is deterministic.  On the other
hand, if the evaluation function is noisy or stochastic, then all 50
individuals should probably be evaluated, so I implement this an
an option.

If an individual undergoes mutation, it is also marked for evaluation,
so the number of evaluations per generation varies depending on the
mutation rate.

Note that it is possible that duplicate individuals are formed
through crossover or mutation, and if the evaluation function is
deterministic and extremely expensive, it might be worthwhile to
identify these duplicates in order to perform the minimal number of
evaluations per generation, but I have never done this.

-- John

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

Date: Fri, 9 Sep 88 12:27:13 EDT
From: John Grefenstette <gref@nrl-aic.arpa>
Subject: 1989 GA Conference

Ken De Jong, who is serving as Conference Chair, has asked me to
announce that the Third International Conference on Genetic
Algorithms and Their Applications will be held June 4-7, 1989,
at George Mason University in Fairfax, VA (just outside Washington, DC).
Dave Schaffer of Philips Labs will serve as Program Chair,
and he will be posting the Call for Papers shortly, with official
deadlines, etc.  But don't delay -- papers will need to be submitted
earlier that in past years, probably in February 1989.

Watch this space for more details.

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

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

