Theory of Evolutionary Algorithms

Summary

The 2004 Dagstuhl seminar on the theory of evolutionary algorithm was the third following seminars in 2000 and 2002 which had the same title. These Dagstuhl seminars are recognized within the evolutionary algorithm community as a biannual series of high quality meetings concerned with theoretical aspects of evolutionary computing. Therefore, there was great interest in participating and presenting ideas. The seminar had 49 participants from twelve different countries, namely eighteen from Germany, eleven from the United States, eight from Great Britain, three from the Netherlands, two from Canada and one each from Austria, Australia, France, Japan, Mexico, Russia, and Switzerland.

From these 49 researchers, 39 were able to give a presentation. The topics of the talks were as diverse as the current status of evolutionary computation theory. There have been talks on general aspects of modeling and analyzing evolutionary algorithms, in particular talks on the pros and cons of infinite population approaches. Also different ways of classifying and analyzing landscapes defined by different problems and genetic operators have been presented. Many talks presented concrete analytical results on specific evolutionary algorithms and specific problems. Such results have been presented for both, continuous and discrete search spaces, for artificial problems or problem classes as well as for a typical combinatorial optimization problem. Recent trends like optimization of noisy objective functions, estimation of distribution algorithms, and coevolution have been topic of several talks as well as the well-established genetic programming. Other randomized search heuristics used for optimization have also been subject of talks, namely estimation of distribution algorithms and ant colony optimization.

The talks have initiated lively and partially controverse discussions. The special Dagstuhl atmosphere has helped a lot to develop a constructive atmosphere that improved mutual understanding and inspiration. The traditional hike on Wednesday afternoon and time spent together in the evenings or late at night let room for more personal discussions, too.

During the seminar the participants were introduced to the new Dagstuhl online proceedings by Jutta Huhse. The idea and advanced possibilities of the new system compared to the old reports was well received. As a result, some of the abstracts here are extended abstracts which are way more detailed than the abstracts of the previous two seminars on the theory of evolutionary algorithms.