17. – 22. Mai 2015, Dagstuhl Seminar 15211
Theory of Evolutionary Algorithms
1 / 3 >
Auskunft zu diesem Dagstuhl Seminar erteilt
Evolutionary algorithms (EAs) are randomized search and optimization methods applicable to problems that may be non-continuous, multi-modal, noisy, multi-objective or dynamic. They have successfully been applied to a wide range of real-world applications and have demonstrated impressive performance in benchmarks for derivative-free optimization. The seminar was devoted to the theory underlying evolutionary algorithms and related methods, in order to gain a better understanding of their properties and to develop new powerful methods in a principled way. The highly international, interdisciplinary seminar brought together leading experts and young researchers in the field. The 45 participants came from 13 different countries, spread over 4 continents. Many additional researchers had expressed their interest to also attend the seminar, but could unfortunately not be considered.
The following report covers all important streams of research in the theory of evolutionary algorithms with a focus on three topics of particular current interest:
Runtime and complexity. Rigorous runtime and analysis and computational complexity theory have become the most important tools in the theory of discrete evolutionary algorithms. The Dagstuhl seminar series ``Theory of Evolutionary Algorithms'' has sparked this development. The drastic increase in new results, new methods, and young researchers entering this field, but also the major unsolved problems naturally lead to keeping this a focus topic.
Information geometry. Using concepts from information geometry in evolutionary algorithms is one of the most promising new theoretical direction in evolutionary computing. The seminar provided a unique opportunity to discuss perspectives and limitations of this approach.
Natural evolution. Evolutionary computing is rooted in theories of natural evolution, and many early approaches to understand basic properties of evolutionary algorithms were inspired by biological evolution theory. Still, today these two research fields are almost completely separated. We invited experts from evolution biology to help better understanding the relations between both fields. We are particularly happy that we succeeded in bringing together researchers from evolution biology and computer science in a way that was stimulating and productive.
The seminar had three types of organized presentation and discussion formats to stimulate the free discussions among the participants. There were 20--30 minutes talks on current topics followed by discussions. These included a talk on potential industrial collaborations. In addition, we had a few longer talks, which combined recent work with an overview over the state-of-the-art in a certain domain: Thomas Jansen spoke on "Understanding Randomised Search Heuristics", Nick Barton on "Limits to Adaptation", Yann Olivier introduced "Information-geometric Optimization", and Timo Kötzing presented a talk on "Stochastic Fitness Functions and Drift". Furthermore, we continued with having "breakout sessions" for longer, parallel group discussions on timely, specialized topics. These were introduced in the last seminar on "Theory of Evolutionary Algorithms". This time, these session were even more productive than previously, both because the organizers and the participants were more used to this format of interaction. The talks and breakout sessions are summarized in Section~4 of this report.
We would like to thank the Dagstuhl team and the attendees for making seminar 15211 a great success and a pleasure to organize.
Creative Commons BY 3.0 Unported license
Benjamin Doerr and Nikolaus Hansen and Christian Igel and Lothar Thiele
Dagstuhl Seminar Series
- 17191: "Theory of Randomized Optimization Heuristics" (2017)
- 13271: "Theory of Evolutionary Algorithms" (2013)
- 10361: "Theory of Evolutionary Algorithms" (2010)
- 08051: "Theory of Evolutionary Algorithms" (2008)
- 06061: "Theory of Evolutionary Algorithms " (2006)
- 04081: "Theory of Evolutionary Algorithms" (2004)
- 02031: "Theory of Evolutionary Algorithms" (2002)
- 00071: "Theory of Evolutionary Algorithms" (2000)
- Data Structures / Algorithms / Complexity
- Optimization / Scheduling
- Soft Computing / Evolutionary Algorithms
- Evolutionary algorithm
- Randomized algorithms
- Theoretical computer science
- Theoretical biology