Optimal sampling strategies for learning a fitness model

摘要

The paper investigates the use of kriging interpolation andestimation as a function approximation tool for the optimization ofcomputationally complex functions. A model of the fitness function isbuilt from a small number of samples of this function. This model isutilized in a model based learning strategy as an auxiliary fitnessfunction. The kriging approach represents a compromise between globalmodels and local models. The model is initially a global approximationof the entire domain, and successive updates during the optimizationprocess transform it into a more precise local approximation. Severalapproaches for the sampling of the true fitness function areinvestigated in order to update a fitness model efficiently and at a lowcomputational cost