Can we make genetic algorithms work in high-dimensionality problems?

Gabriel Alvarez

gabriel@sep.stanford.edu

ABSTRACT

In this paper I compare the performance of a standard genetic algorithm versus a micro-genetic algorithm for matching a randomly-generated seismic trace to a reference trace with the same frequency spectrum. A micro genetic algorithm evolves a very small population that must be restarted whenever the population loses its genetic diversity. I show that the micro-genetic algorithm is more efficient in solving this problem in terms of improved rate of converge, especially in the first few generations. This characteristic may make the method useful for locating the most promising valleys in the search space which can then be searched with more traditional gradient-based methods. An additional benefit is a significant reduction in the number of evolution parameters that needs to be adjusted making the method more easy to use.