Bayesian Optimization Algorithm, Population Sizing, and Time to Convergence
Martin Pelikán, Martin Pelikán, Erick Cantú‐Paz, David E. Goldberg, David E. Goldberg · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 2000
This paper analyzes convergence properties of the Bayesian optimization algorithm (BOA). It settles the BOA into the framework of problem decomposition used frequently in order to model and understand the behavior of simple genetic algorithms. The growth of the population size and the number of generations until convergence with respect to the size of a problem is theoretically analyzed. The theoretical results are supported by a number of experiments. 1 Introduction Recently, the Bayesian optimization algorithm (BOA) has proven to optimize problems of bounded difficulty quickly, reliably, and accurately. The number of function evaluations until convergence was investigated on a number of problems. However, the questions of (1) how to choose an adequate population size in order to solve a given problem with a certain degree of accuracy and reliability, and (2) how many generations it will take until the algorithm converges, remained unanswered. This paper makes an important step...