Improving the Performance of Genetic Algorithms through Derandomization

Genetic algorithms are believed by some to be very eecient optimization and adaptation tools. So far, the eecacy of genetic algorithms has been described by empirical results, and yet theoretical approaches are far behind. This paper aims at raising fundamental theoretical questions about the utility and performance of genetic algorithms from an algorithmic point of view, i.e., space requirements and computational complexity. These questions originate from various existing theories and the no-free-lunch theorem, which compares all possible optimization procedures with respect to an equal distribution of all possible objective functions. While these questions are open at least in part, they all strongly indicate that genetic algorithms are less eecient than other (determinis-tic) optimization algorithms. From an algorithmic point of view, the no-free-lunch theorem suggests that the random application of variation operators signiicantly degrades the performance of genetic algorithms , since resampling of already visited points is not avoided. Consequently, this paper proposes a simple transformation that is strictly more eecient than the equivalent genetic algorithm.