Complex Systems

Convergence in Simulated Evolution Algorithms Download PDF

Olivier François
Laboratoire de Modélisation et Calcul,
Institut IMAG, Université de Grenoble,
BP 53, 38041 Grenoble cedex 9, France


New simulated evolution algorithms are presented for global minimization. The main feature of these algorithms is to couple the standard mutation and selection operators into a single operator. Exponential probabilities are also introduced so as to stress the relevance of statistical mechanics arguments. The work follows the stream of large deviations techniques developed to analyze the simulated annealing and genetic algorithms.