This talk starts with a brief outline how to transfer the convergence velocity analysis from evolution strategies to genetic algorithms.
The focus of the talk deals with the principle of self-adaptation in evolutionary algorithms, which provides the fundamental parameter control method exploited by evolution strategies. After explaining the basic idea and classifying existing parameter control mechanisms that have been suggested in evolutionary algorithm research, the evolution strategy method for self-adaptation is explained in detail. This presentation includes both empirical results and recent theoretical investigations which clarify the robustness of the self-adaptation principle to work under a variety of algorithmic conditions. Other attempts to self-adapt mutation rates in genetic algorithms or recombination operators are also briefly explained.
Based on our experience with all instances of evolutionary
algorithms, we summarize our general point of view on their
usefulness and advantages as well as disadvantages and conclude
the talk by mentioning the industrial application problems which
are currently tackled at the Center for Applied Systems Analysis
(CASA) with various (parallel) variants of evolutionary algorithms.
Back to Workshop Schedule