Optimal Simulator Selection

Computer simulators are widely used for the study of complex systems. In many applications, there multiple available with different scientific interpretations underlying mechanism, and goal is to identify an optimal simulator based on observed physical experiments. To achieve goal, we propose a selection criterion leave-one-out cross-validation. This consists goodness-of-fit measure generalized degrees freedom penalizing sensitivity perturbations in observations. Asymptotic properties selected discussed. It shown that proposed procedure includes conventional calibration method as special case. The finite sample performance demonstrated through numerical examples. application cell biology, selected, which can shed light T recognition mechanism human immune system. Supplementary materials this article online.