Bayesian Methods in Global Optimization
Bruno Betrò · 1992
The basic problem in global optimization is the one of properly assessing a performance criterion for global optimum seeking algorithms, because of the lack of manageable analytical characterizations of the global optimum. In order to cope with this difficulty, the idea has been introduced of superimposing to the global optimization problem a probabilistic structure and to set up accuracy criteria consequently. Then Bayes theorem provides the basic tool for adapting the probabilistic structure to information gained about the problem itself after a certain number of function evaluations. Methods derived within this framework will be referred to as Bayesian methods. Following [1], an overview is presented of different probabilistic formulations of the global optimization problem and of related Bayesian methods as yet proposed.