Department of Computer Science
Chair of Algorithm Engineering (Ls11)
Home Contact Deutsch English
menu-en
Learning from Failures in Evolutionary Computation - GECCO 2009 workshop

Learning from Failures in Evolutionary Computation

Workshop at GECCO 2009

Organization: Nicola Beume and Mike Preuss





Vision

No experiment is ever a complete failure, it can always serve as a bad example. From a failed study we cannot exactly learn what we intended to but we can always learn something. This shall be the mission statement of this workshop.

Failures frequently happen in experimental as well as in theoretical research. The question is what we can learn from a specific failure, and on a higher level, how to deal with these failures. The main emphasis is thus twofold: To inform about concrete failed approaches, and to review and develop methods of attaining progress beyond the failures.

Participants will present seemingly clever ideas and concepts that somehow did not work. These can e.g. be the beginning of the train of thoughts which already led to another successful approach, or it may be the presentation of a problem which is still unsolved. Sharing experiences on failures will help understanding e.g. a certain method and its area of application, a certain problem, or the relation between an approach and a problem. Learning that a certain method for theoretical analysis did not provide the desired insights may give an impression for which kind of problems and questions it is suitable. For experimental works, there is a strengthening movement of refining the methodology with regard to parameter analysis and enhanced statistics.

It is intended that differences and similarities of the current scientific approaches of several researchers are discussed, as well as the possibilities of 'negative papers'. This relates to many issues in the philosophy of science and may contribute to a further development of experimental and theoretical methodologies in evolutionary computation.



top


Schedule

The workshop takes place at Rez-de-Chaussee / Lobby Level, Verriere A on Wednesday, July 8, 2009.

Download Overview of Workshop Program.
Download Program with Abstracts of Talks.

14:00 - 14:10 Opening by Organizers
14:10 - 15:00 Failures as Stepping Stones to Success or "per aspera ad astra"
Hans-Paul Schwefel (Invited Speaker)
15:00 - 15:25 On the Limitations of Adaptive Resampling using the Student's t-test in Evolution Strategies
Johannes W. Kruisselbrink, Michael T.M. Emmerich, Thomas Bäck
15:25 - 15:50 Reinforcement Learning for Games: Failures and Successes
Wolfgang Konen, Thomas Bartz-Beielstein
15:50 - 16:10 Break
16:10 - 17:00 Compete to Win, Lose to Learn - Experiences from Game Competitions
Pier Luca Lanzi (Invited Speaker)
17:00 - 17:25 A Series of Failed and Partially Successful Fitness Functions for Evolving Spiking Neural Networks
J. David Schaffer, Heike Sichtig, Craig Laramee
17:25 - 17:50 Lessons Learned in Evolutionary Computation: 11 Steps to Success
Jörn Mehnen
17:50 - 18:00 General Discussion and Closing by Organizers


top


Invited Speakers

Hans-Paul Schwefel, TU Dortmund University
Failures as Stepping Stones to Success or "per aspera ad astra"

Abstract. The implicit thesis of this talk's title will be underpinned with some examples from (my) real life. A first example leads back to the 1960s, when I simulated the (1+1)-ES with discrete mutations on a two-dimensional parabolic ridge by means of a Z23 computer. The result - getting stuck in certain search directions - led to making use of gaussian variations.
The second example comes from experimental investigations to determine the shape of a hot water flashing nozzle, the water being really hot and not simulated on a computer. In search for a multimembered evolutionary algorithm with effective self-adaptation of the mutation strengths, a couple of failures occurred. These, however, rendered deep insight into basic prerequisites to achieve the goal. And finally, some theory will be re-presented about the optimal failure rate in two black-box situations.



Pier Luca Lanzi, Politecnico di Milano
Compete to Win, Lose to Learn - Experiences from Game Competitions




top


Accepted Papers

Reinforcement Learning for Games: Failures and Successes
Wolfgang Konen, Thomas Bartz-Beielstein

On the Limitations of Adaptive Resampling in Evolution Strategies using the Student's t-test in Evolution Strategies
Johannes W. Kruisselbrink, Michael T.M. Emmerich, Thomas Bäck

Lessons Learned in Evolutionary Computation: 11 Steps to Success
Jörn Mehnen

A series of failed and partially successful fitness functions for evolving spiking neural networks
J. David Schaffer, Heike Sichtig, Craig Laramee



top


Topics (roughly)

  • Concrete failures in experimental investigations
  • Concrete failures in theoretical investigations
  • Possibilities of 'negative papers'
  • General methodological issues


top


Submission

We invite submissions in the range of short descriptions of an interesting enlightening failure up to full papers.
Please send your submissions (2-8 pages in ACM style) to: lffec-sub@ls11.cs.tu-dortmund.de.
Accepted papers will be published in the GECCO 2009 companion material and be included with the proceedings on a CD, and also in the ACM Digital library.
Submission closed.



top


Important Dates

Submissions due: April 3, 2009 (extended)
Decision notifications: April 10, 2009
Camera ready version due: April 17, 2009
Registration for presenting authors due: April 27, 2009
GECCO 2009 conference: July 8-12, 2009
Workshop: July 8, 14:00 - 18:00



top


Program Committee

Carlos M. Fonseca, Universidade do Algarve
Thomas Jansen, University College Cork
Thomas Stützle, Université Libre de Bruxelles



top


Insipiring Citations and Links

Science must begin with myths, and with the criticism of myths.
Karl Popper.

Whenever a theory appears to you as the only possible one, take this as a sign that you have neither understood the theory nor the problem which it was intended to solve.
Karl Popper.

With engineering, I view this year's failure as next year's opportunity to try it again. Failures are not something to be avoided. You want to have them happen as quickly as you can so you can make progress rapidly.
Gordon Moore

Failure is only the opportunity to begin again more intelligently.
Henry Ford.

There is no failure. Only feedback.
Robert Allen.

I have not failed. I've just found 10,000 ways that won't work.
Thomas Edison.

Try again. Fail again. Fail better.
Samuel Beckett.

Engineering Disasters and Learning from Failures

Imprint
<webmaster  ls11.cs.tu-dortmund.de>
The university does not accept liability for the contents of linked external internet sites