Using a double-based genetic algorithm on a population of computer programs

In this paper, we present a new approach, which improves the performance of a genetic algorithm. Genetic algorithms are iterative search procedures based on natural genetic. We use an original genetic algorithm that manipulates pairs of twins in its population: DGA, double-based genetic algorithm. We show that this approach is relevant for genetic programming, which manipulates populations of trees. In particular, we show that doubles enable to transform a deceptive problem into a convergent one. We also prove that using pairs of double functions in the primitive function set is more efficient in the problem of learning boolean functions