Towards self-adapting evolution strategies

Optimization algorithms imitating certain principles of nature have proved their capability in various domains of applications. Dealing with parameter optimization problems one usually trades the original problem for a much simpler one, estimating the exogenous parameters of the algorithm chosen to yield a good solution as fast as possible. On the one hand, this paper demonstrates empirically for a small set of test functions, how convergence velocity and reliability of evolution strategies depend on the recombination operator chosen. On the other hand, first results indicate that the capability of self-adaptation within evolution strategies may be exploited in order to reduce the number of exogenous parameters, thus leading to more robust strategies