A modified random implementation of the level-value estimation method for global optimization

摘要

In this paper, we propose a modified random implementation of the level-value estimation method for global optimization which was proposed in [11]. In the modified implementable algorithm, we employ a kernel density function in nonparametric function estimation as the importance sampling density, and use the main idea of the cross-entropy method to update this density function. Numerical results show that the modified implementation is better numerical precision than that one which uses a coordinate-normal density as the sample density.