A pseudo-Bayesian method of global optimization

A method of global searching which takes some of the advantageous principles of Bayesian methods such as memory of past evaluations, yet also uses principles of genetic algorithms such as parallel structure and reduced complexity. is discussed. Results for this method are found on the basis of the number of evaluations needed to converge upon the global solution for a standard test function. The algorithm is shown to converge probabilistically as the number of evaluations approaches infinity, and is shown to have a computational complexity of O(i), where i is the number of iterations.<>