Uncertainty in Surrogate Models

CCS Concepts •Computing methodologies → Continuous space search; The existing infill sampling criteria for updating surrogate models come from two objectives: the optimums and the most uncertain points of the surrogate model. The optimums of the surrogate model can improve the exploitation ability of the surrogate model, while most uncertain solutions can help improve the exploration ability by searching the unexplored regions and most efficiently enhancing the accuracy of the surrogate model. It has been shown that the efficiency of surrogate models can be raised if the uncertainty information on the surrogate models is properly used. In this presentation, three points are discussed.