A Boltzmann based estimation of distribution algorithm

The Elitist Convergent Estimation of Distribution Algorithm (ECEDA), is a definition of a class of EDA which guarantees convergence to the optimum. This paper introduces the conceptual ECEDA and a practical approach derived from it, called the Boltzmann Univariate Marginal Distribution Algorithm (BUMDA). The BUMDA uses a Gaussian model to approximate the Boltzmann distribution, requiring only one user given parameter: the population size. Several experiments and statistical analysis are used to contrast the BUMDA with state of the art EDAs.