| Popis: |
The search for traveltime parameters is a global optimization problem. Several metaheuristics have been proposed to locate the global optima to compute the least amount of their objective functions. However, the theoretical limitations imposed by the no-free-lunch theorem restrict the optimality of such metaheuristics. To escape those limitations, we propose a coevolutionary approach called evolution by neighborhood similarity, which is outside the scope of the restrictions of this theorem and allows us to speed up the search convergence. The technique's effectiveness is based on the approach to exchanging the best individuals found between suitable domains during a differential evolution metaheuristic execution. Moreover, we further expand our technique to graphics processing units, allowing us to explore the performance and memory aspects of the method. Ultimately, our complete coevolutionary algorithm can speed up the parameter search by more than five times and reduce the energy consumption by more than thirty-three times compared to a regular metaheuristic implementation. Although some overheads are still problematic in the technique, we present a first approach that effectively explores the data redundancy hidden in the estimation process, allowing us to provide qualitative, performance, and scalable improvements. |
