Genetic Algorithms Digest    Thursday, 22 February 1989    Volume 3 : Issue 5

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

Today's Topics:
	- Moderator's Note
	- Training neural nets with GAs
	- Neural Nets and GAs, and Avaliable Papers
	- 2-Armed Bandits (2)

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

Date: Thu, 22 Feb 89 9:15:04 EST
From: ga-list-request@aic.nrl.navy.mil (John J. Grefenstette)
Subject: Moderator's Note

Several readers have written wondering what happened to GA-List.
There has been a marked decrease in traffic this month, probably
because people were busy writing up their results for ICGA-89.
Now that the submission deadline has passed, I expect
things will pick up again.

- John

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

Date: Tue, 7 Feb 89 18:48 EST
From: DMONTANA%COOPER@rcca.bbn.com
Subject: Training neural nets with GAs

Due to a reduction of secretarial staff from 3 to 1 and the large
number of requests, there was some confusion in sending out copies
of the paper "Training Feedforward Neural Networks Using Genetic
Algorithms".  Anybody who has not received a copy should ask me again
at dmontana@bbn.com.  If you mention that you had previously requested
a copy, I'll put you at the front of the list.  Sorry for any mixups.

Dave

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

Date: Thu, 9 Feb 89 10:41:14 MST
From: whitley@ravel.CS.ColoState.Edu (darrell  whitley)
Subject: Neural Nets and GAs, and Avaliable Papers

This posting contains two items.

First, I am gathering information about user interface 
tools for neural nets and genetic algorithms. We are 
building an x-windows package that allows one to build
neural nets, run genetic algorithms, back propagation, etc.
I am getting help from a Human Interface team at NCR.
Before we get deeper into this, however, we want to look
at existing user interface tools.  I have "Genesis" by 
John Grefenstette for genetic algorithms and  the Rochester
Simulator for neural nets. Please sent mail about any other
systems which you think we should be aware of. We are 
especially interested in public domain software, but want 
to know about any and all such tools.    Our tool will be
public domain and available to the readers of this net.


Second, we have a number of papers that can be obtained
by sending us email:
    
Whitley D. (1989) 
Applying Genetic Algorithms to Neural Network
Learning.
To appear in:
Proc. 7th Conf. for the Study of 
Artificial Intelligence and Simulated Behavior.
Sussex, England:  Pitman Publishing.

Whitley D., and Starkweather, T.  (1989)
Genetic Algorithm Applications: 
Neural Nets, Traveling Salesmen and Schedules.
To appear in:
1989 Rocky Mountain Conference on Artificial Intelligence
Denver, CO. 

Whitley D.  (1989)
Optimizing Neural Networks Using Genetic Algorithms.
Special Neurocomputing Issue of Design&Elektronik.
Markt&Technik, Publishers, Munich, Germany.

Whitley D., and Hanson, T. (1989)
Optimizing Neural Networks Using
Faster, More Accurate Genetic Search.
(This is a longer version of two papers we have submitted 
to conferences.  It summarizes most of the work we have
done in the past year.  It also explains how we increased
genetic search speed by more than 10 fold and achieved MORE ACCURATE
solutions, typically 10e-10, on larger problems.)

Whitley (1989)
The GENITOR algorithm and Selective Pressure
OR
Why Rank Based Allocation of Reproductive Trials is Best
(Soon to appear under a Tech Report Number.) 

We have other papers on Genetic Algorithms we have
just finished, but these should be of the most 
general interest.
We have a new recombination operator for the Traveling 
Salesman Problem that has an underlying binary
representation, requires no "correction" in the offspring
it produces, and has produced new "best known" results 
on 50 city and 75 city problems.  We have used the same
operator for sequencing jobs in a job shop schedule.
We also have a new schema theorem that uses the assumption
of bit-at-a-time recombination in order to make predictions
about string gains.  The result is predicted increases of above
average schemata, even when the fitness ratio is relatively
low (eg 1.05).

Back to the original topic:  if you have pointers
to graphics tools / user interfaces for neural nets
or genetic algorithms please let me know.

   Darrell Whitley       whitley@ravel.cs.colostate.edu

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

Date:     15-FEB-1989 09:34:58
From: <75008378%VAX2.NIHED.IE@CUNYVM.CUNY.EDU>
Subject: 2-Armed Bandits

In Holland's "Adaptation in Natural and Artificial Systems" there is a
discussion and analysis of the "optimal allocation of trials", and the
so-called 2-armed bandit in particular.  I'm interested in this problem,
but am finding it difficult to grasp some aspects of Holland's analysis.
I'm chasing his original references on the subject, but I'd also
appreciate knowing of any other relevant material - particularly any
more recent commentary on Holland's original work that might give me another
avenue into it.

Thanks,

Barry McMullin, NIHED, Dublin, IRELAND.
<MCMULLINB@VAX2.NIHED.IE>

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

Date: Thu, 16 Feb 89 09:33:17 EST
From: John Grefenstette <gref@AIC.NRL.NAVY.MIL>
Subject: Re: 2-Armed Bandits

James Baker and I have submitted a paper on this topic to ICGA-89.
The paper examines, among other things, how well the analogy with
k-armed bandits describes the behavior of genetic algorithms in practice.
We think that there many be better ways to analyze the behavior
of genetic algorithms, and offer some suggestions along those lines.
We'll be happy to make the paper available, after the reviewers get
through with it.

- John

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

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

