Prashant Singh
Division of Scientific Computing
Department of Information Technology
Uppsala University
Uppsala
Numerous real-world optimization problems often involve expensive-to-evaluate objective functions. Problems are compounded if the optimization problem consists of searching for a trade-off between multiple objectives. Traditional methods of solving such problems require a large number of objective function evaluations. This translates into a very slow and lengthy optimization process that may be impractical for the problem at hand. Surrogate modeling is a viable alternative that consists of training a cheap-to-evaluate approximation of the expensive objective function. The surrogate model offers instant evaluation and can be used in lieu of the objective function. Surrogate-based optimization takes the use of surrogate models a step further and makes use of model predictions and statistical sampling techniques to solve the optimization problem while minimizing the number of evaluations of the objective function.