Metaevolution: Synthesis of Optimization Algorithms by Means of Symbolic Regression and Evolutionary Algorithms

This thesis is aimed at the explanation as to how Analytic Programming could be used for the synthesis of new optimizing algorithms, probably of evolutionary character. Evolutionary algorithms are tools for the optimization of difficult tasks. The principle of this thesis is to show that it might be possible to synthesize a powerful algorithm based on evolutionary ideas. The name of this thesis ? metaevolution ? covers all these ideas. Metaevolution is, according to previous approaches, determining the optimal evolutionary algorithm, best types of evolutionary operator and their parameter setting for a given problem. It means basically, that one evolutionary algorithm tunes another one. But this approach is novel. We use metaevolution for synthesis a new algorithm completely, not only for setting of its parameters. The book shows different applications with AP and simulations with new synthesized algorithms. The results are arranged in tables and charts to present the robustness of the method.