Dynamics of Evolutionary Algorithms

Discussion Report

The 2001 Genetic and Evolutionary Computation Conference (GECCO-2001)

July 8, 2001, 2:00pm - 5:30pm, San Francisco, California, USA


This page selectively highlights some of the discussions. Particularly those after the coffee break.


Prior to the Coffee Break things ran pretty much as planned.
The workshop was a roundtable open discussion, moderated by W. B. Langdon, where key proponents of the different approaches to theory listed below gave short talks in support of their approach (and against others). Following the coffee break less formal discussions were held in the round.

Michael Vose Microscopic models - Markov chains.

Kalyanmoy Deb Approximate-models-that-work

Lashon Booker Inspiration from Population biology.

Chris Stephens Coarse graining (schema-based) approaches

Dirk Thierens felt that the goal of GA theory should be practical and therefore scalable genetic algorithms. Experimental analysis is important to test and guide development of theory.

Other Points

Various points were offered for discussion. Including Jim Smith who is currently focussed on developing models of self-adaptive and memetic algorithms, and refining landscape analysis and generation techniques. James Napier (American Express) is interested in using GA theory in practise. Terry Soule is interested in neighbourhood operators, the dynamics of the effective fitness landscape and tools that might give helpful insight, such as crossover correlation coefficients. Anthony Liekens (Einhoven) expressed interest in diploid chromosomes. While Edwin de Jong is interested in predicting recombinative systems.

Steve Gustafson (University of Nottingham) is interested in the parallel between the struggle in biology to ``prove'' evolution and that in evolutionary computation to prove some theory or ``schema theorem'', especially in genetic programming.

Riccardo Poli said that new Schema Theorems are true equalities which allow one to predict exactly the expected characteristics of the population at the next time step. As a result exact schema theorems can also be used backwards through time. I.e. to find under which conditions at the previous generations a GA or GP system can be expected to be in a given state at a particular generation with a known probability.

While neither Jon Shapiro nor Adam Prugel-Bennett were present, there was a brief discussion on their statistical thermodynamics approach to genetic algorithm theory.


W. B. LangdonW.Langdon@cs.ucl.ac.uk
Terry Souletsoule@cs.uidaho.edu
Edwin de Jongedwin@cs.brandeis.edu
Jon Rowej.e.rowe@cs.bham.ac.uk
R. Polir.poli@cs.bham.ac.uk
Anthony Liekensmooby@alife.org
Chris Stephensstephens@nuclecu.unam.mx
Larry MerkleLarry.Merkle@usafa.af.mil
Alden Wrightwright@cs.umt.edu
James A. NapierJames.A.Napier@aexp.com
Nic McPheemcphee@mrs.umn.edu
Kalyanmoy Debdeb@iitk.ac.in
Jim SmithJames.Smith@uwe.ac.uk
Hideaki SuzukiHSuzuki@isd.atr.co.jp
Steven Gustafsonsmg@cs.nott.ac.uk
Lashon Bookerbooker@mitre.org
Dirk Thierensdirk@cs.uu.nl
Ken De Jongkdejong@gmu.edu
Michael D. Vosevose@cs.utk.edu

What Next?

Theoretical insights into evolutionary algorithms will doubtless be discussed at many events in the coming months. However the following spring to mind: May be a different approaches are need?
May be a more formal record of these discussions is needed?


Christopher Stephens
Instituto de Ciencias Nucleares
Circuito Exterior
A. Postal 70-543
Mexico DF 04510
Email: stephens@nuclecu.unam.mx

Riccardo Poli
School of Computer Science
The University of Birmingham
Birmingham B15 2TT

W.B.Langdon $Revision: 1.10 $ $Date: 2011/05/10 17:45:39 $