The proposed research will involve designing, analyzing, and testing algorithms for computer simulation experiment design and analysis. The objective is to improve the computational efficiency of simulation experiments in settings where there is a sequence of repeated experiments using the same simulation model with different inputs. The proposed algorithms will improve efficiency of later experiments by storing and reusing the results of earlier experiments. In contrast, standard practice is to perform and analyze every experiment in the sequence separately, throwing the results of one experiment away as soon as it delivers an answer. The methods to be employed to reuse the results of experiments include simulation metamodeling and variance reduction techniques for stochastic simulation. In designing and analyzing the resulting algorithms, we will leverage existing knowledge on simulation experiment design and analysis, including optimal design of experiments, metamodel validation, and adaptive design of sequential experiments. The proposed algorithms will be tested in computer experiments using examples, which are representative of realistic applications, of sequences of simulation experiments using the same simulation model but different inputs. In each example, the efficiency gain generated by the proposed algorithms, compared to standard practice, will be measured in terms of reduced computational cost given a target accuracy, or in terms of increased accuracy given a fixed computational budget.
|Effective start/end date||1/1/17 → 6/30/21|
- National Science Foundation (CMMI-1634982)