
Genetic Algorithms Digest   Monday, 10 December 1990   Volume 4 : Issue 27

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

Today's Topics:
	- Messages from moderator
	- MIMD Parallel classifier systems
	- Reals instead of Gray code (Re: GA workshop abstract)
	- Re: good questions on Evolvability
	- concept learning with GAs
	- Symposium announcement

******************************************************************************

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

4th Intl. Conference on Genetic Algorithms (v4n17)            Jul 14-17, 1991
AAAI91 - Deadline for submissions is Jan 30, 1991
Machine Learning Conference - Deadline for submissions is Feb 1, 1991

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

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

Date: 5 November, 1990
From: Alan C. Schultz (GA-List moderator)
Subject: We're back.

	Due to some crunches on our end, there has been a gap in the 
	usually regular delivery of ga-list.  However, we have not been
	receiving many submissions lately.  Without submissions, there
	is nothing to send out.  So lets get some discussions going!

	Last issue I promised more information on Dave Davis'
	forthcoming book on GAs.  I don't have it yet.  NEXT issue
	(I promise) there will be more information, including ordering
	info. 
	
	Alan C. Schultz

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

Date: Mon, 12 Nov 90 18:15:38 PST
From: Geoffrey Ballinger <aighb@castle.edinburgh.ac.uk>
Subject: MIMD Parallel classifier systems

	   I have read several articles about SIMD implementations of
   classifier systems but I haven't come across any about MIMD
   implementations. I am just starting a project to design and implement an
   MIMD based classifier system and I was wondering if there has been any
   work done in this area (to save me reinventing to many wheels!)? Thanks,

			   Geoff.

    Geoff Ballinger,                       JANET: Geoff@Uk.Ac.Ed
    Department of Artificial Intelligence, UUCP: ...!mcsun!ukc!Ed.Ac.Uk!Geoff
    Edinburgh University.                 

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

Date: Mon, 12 Nov 90 18:15:38 PST
From: autodesk!thezahir!peb@uunet.UU.NET (Paul Baclaski)
Subject: Reals instead of Gray code (Re: GA workshop abstract)

     Please see "A New Interpretation of Schema Notation That Overturns
     the Binary Encoding Constraint" by Jim Antonisse, Proceedings
     of the 3rd Internation Conference on Genetic Algorithms.  I don't 
     know of any later references to this subject.

     I read this after I too used real numbers in my GAs.  I used them
     simply because I thought Gray codes would be too inefficient in 
     the language I was using (Scheme).  Some other benefits can result
     from such an encoding, e.g., mutation rate could be customized for
     each Real gene.

     Paul E. Baclaski
     peb@autodesk.com

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

Date: Thu, 29 Nov 90 21:43:42 -0600
From: William Mike Rudnick <rudnick@ux1.cso.uiuc.edu>
Subject: Re: good questions on Evolvability

     Yes, the question you raise and call "evolvability" is central to the
     use of GAs.  To paraphrase and generalize: "When will the GA work and
     when won't it work?"  Good question.  My reading of the GA literature
     suggests we don't know the answer yet, but we're working on it and
     making some progress, eg, the recognition of classes of GA
     problem-encodings/embeddings (deceptive, partially deceptive,
     non-deceptive) and the recognition of sampling bias/error as
     exemplified in the needle-in-the-haystack problem.

     About noise in the objective function -- again you raise a very good
     question.  Most of the work done has been empirical, with only (to my
     knowledge) John Grefenstette and Michael Fitzpatric's (ICGA '85)
     picture registration work including some analysis.  Yuval Davidor's
     recent work on epistasis (FOGA '90 and to appear in Complex Systems)
     also bears on noise in GAs, I believe.  I've recently begun working on
     some aspects of the GA-noise problem.  It turns out there are a number
     of kinds of potential noise, or things that can sensibly be viewed as
     noise, in GA function.  Objective function noise is one of these.  I
     have the impression the GA community tends to view GAs as robust in
     the presense of noise, but this is yet to be proven or even quantified
     in any general sense.

     Comments, anyone?

       Mike Rudnick

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

Date: Sat, 1 Dec 90 15:20:19 -0600
From: Riyaz Sikora <sikora@uxh.cso.uiuc.edu>
Subject: concept learning with GAs

     For those of you interested in concept learning using GAs,
     the following technical report is now available:

     BEBR Faculty Working Paper No. 90-1693.

     "A Double-Layered Learning Approach to Acquiring Rules for Financial
     Classification"

			     ABSTRACT
      In this paper we describe a machine learning approach, based on a
     double-layered architecture and the Genetic Algorithm (GA), to learning
     decision rules for financial classification. These rules can be
     implemented in an expert system for future consultation in the
     particular application area. GAs represent a new class of learning
     algorithms based on the model of the biological evolution process.
     Equipped with unique search behaviour and solution-seeking properties,
     GAs provide an interesting new technique for such
     financial-classification tasks as bankruptcy prediction and credit
     analysis. However, some modifications to the basic GA would be necessary
     in order to make the method suitable for solving the
     financial-classification problems. One of the objectives of this paper
     is to identify the aspects of GAs that need to be modified for the
     classification domain and how they can best be incorporated in the GAs.
     More importantly, we expand on the concept of GA and develop a learning
     method called Double-layered Learning System (DLS) that integrates GA
     with a similarity-based learning technique called Probabilistic Learning
     System (PLS1). DLS proves to be an effective improvement over both GA
     and PLS1. An analysis is included to evaluate the performance of DLS in
     terms of computational efficiency, prediction accuracy, conciseness of
     the concepts generated, and rule refinement.

     To get a hard copy of the above report please send a request to
	     sikora@uxh.cso.uiuc.edu
     or write to:
	     Bureau of Economic and Business Research,
	     University of Illinois, Urbana-Champaign,
	     428 Commerce West,
	     1206 South Sixth Street,
	     Champaign, IL-61820.

     Riyaz Sikora.

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

Date: Tue, 6 Nov 90 08:16:24 EST
From: schultz@AIC.NRL.Navy.Mil
Subject: Forwarded Symposium announcement

			      CALL FOR PAPERS

			       A SYMPOSIUM ON

			   ASSOCIATE TECHNOLOGY:
	    PLANNING AND DECISION SUPPORT IN EVOLVING SITUATIONS

			   DATE: JUNE 6 & 7, 1991

		     LOCATION: GEORGE MASON UNIVERSITY
			       FAIRFAX, VIRGINIA    


				SPONSORED BY

		 DEFENSE ADVANCED RESEARCH PROJECTS AGENCY
			     BDM, INTERNATIONAL
			  GEORGE MASON UNIVERSITY 

     With the advent of programs such as Pilot's Associate, the 
     Rotorcraft Pilot's Associate, SOAS and several others there has 
     emerged an increasing interest in Associate Technology - the 
     development of systems to support real-time planning and decision 
     making in dynamic and evolving situations (e.g., systems that 
     support on-board mission planning and replanning in a pilot 
     cockpit).  At the same time, other research communities are ex-
     ploring issues very related to associate technology.  For in-
     stance, real time planning and the problem of interleaving plan-
     ning and execution have recently emerged as key research areas in 
     AI.  Similarly, decision making in dynamic situations is gaining 
     increasing attention in the cognitive psychology community.

     This symposium has two objectives.  First, to assess the current 
     state of the science and appropriate research directions in as-
     sociate technology.  Second, to explore the relationships and 
     synergies between research areas relevant to associate technol-
     ogy.  

     Submissions

     Researchers who wish to present at this symposium are invited to 
     submit four (4) copies of an extended abstract or completed paper 
     to:
	       Paul E. Lehner
	       Information Systems and Systems Engineering
	       George Mason University
	       Fairfax, Virginia   22030

     Completed papers should be no more than 12 single spaced pages, 
     including figures, tables and references.

     Topics of Interest

     We are interested in a broad spectrum of papers relevant to the 
     topic of associate technology.  University, commercial and 
     government sources are encouraged to submit.  Both technology and 
     application oriented papers are of interest.  Some example topics 
     are listed below.

	  Architectures for associate technology systems
	    description of current systems
	    requirements for future systems
	    potential application areas
	    mission planning and replanning

	  Automated Reasoning
	    real time inference and planning
	    probabilistic reasoning in dynamic situations
	    interleaving planning and execution

	  Human Decision Making
	    inference in time-constrained/evolving situations
	    dynamic decision making
	    team decision making

	  Human/Computer Interaction
	    operator modeling
	    pilot/vehicle interface


     Deadlines

       Submission of Extended Abstract  --  February 23, 1991
       Notification of Acceptance       --  March 15, 91
       Receipt of Completed Paper       --  April 30, 1991


     For further information, please contact one of the following:

	  Paul E. Lehner                Phil Merkel
	  Information Systems and       BDM International, Inc.
	       Systems Engineering      Suite 750
	  George Mason University       4001 North Fairfax Drive
	  4400 University Drive         Arlington, Virginia  22203
	  Fairfax, Virginia  22030      (703) 247-0357
	  (703) 323-4355                merkel@a.isi.edu
	  plehner@gmuvax2.gmu.edu 

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