Theory of Evolutionary Algorithms

During the last ten years computers have become fast enough to support evolutionary algorithms and a lot of applications to real-world problems have been developed. This has led to a great deal of empirical knowledge on the behavior of evolutionary algorithms and to many heuristics for choosing their associated parameters. There is also a developing theory of evolutionary algorithms based on tools from the analysis of randomized algorithms, of Markov processes, and of dynamical systems. The aim of this workshop was to contribute to this theory and to allow a discussion between researchers with different backgrounds. The organizers are happy to report that 45 researchers accepted an invitation to Dagstuhl. They came from Germany (21), England (8), USA (5), France (2), Netherlands (2), Romania (2), Austria (1), Belgium (1), India (1), Mexico (1), and Poland (1). The 31 talks captured all the aspects of a theory of evolutionary algorithms, among them statistical dynamics, time-varying landscapes, convergence issues, complexity results, fitness landscapes, models of evolutionary algorithms, analysis of the run time of evolutionary algorithms, self-adaptation, new variants of evolutionary algorithms, and genetic programming. The discussion was extremely vivid. There was almost no talk that evoked fewer than five questions and remarks. The schedule included an informal evening session where eight topics suggested by the participants were discussed. Besides the official schedule the participants used unscheduled time for many discussions and some informal sessions with short talks, all inspired by the special Dagstuhl atmosphere. The special event of the week was the Wednesday hike where it has snowed heavily on the way out and the sun shone on the way back through the snow.