Statistical Methods in Computational Modeling and Virtual Experiments

点击翻译按钮获取中文摘要

PI: John Tucker Proposal Number: DMS 9714380 Project: Statistical Methods in Computational Modeling and Virtual Experiments Abstract: The Committee on Applied and Theoretical Statistics (CATS) will convene a multidisciplinary study panel on statistical methods in computational modeling and virtual experiments. The panel will recommend sound methods for identifying and quantifying uncertainties that arise in simulation experiments. The panel will develop recommendations for improved and, where necessary, new statistical methods and data analysis techniques for use in simulation modeling and virtual experiments. It will also recommend appropriate statistical methods to assist practitioners in their use of experimental data when validating simulation models or assessing model fit and predictive performance. These recommendations will address modeling and simulation limitations associated with ever increasingly complex systems. The panel's report will provide examples to illustrate the complexity of evaluating model combinations that arise from varying assumptions, parameters, and initial conditions within the models, will point out possible approaches for addressing those complexities from existing experimental design techniques, and will highlight areas needing further research. Experts may be consulted from the communities concerned with computer simulation, discrete events systems, and dynamic systems modeling, and any other scientists, researchers, or engineers with significant technology or methodology to contribute. A National Research Council panel of experts will examine the state-of-the-art for statistical methods used in computer simulation and modeling, and computational experimentation. This is needed because of an ever-growing dependence upon these kinds of simulations and "virtual" experiments in designing and evaluating such complicated devices and systems as semiconductors, automobiles, aircraft, spacecraft, or in very complex application areas such as transportation networks, government intelligence, and battlefield scenarios. The study will address how modeling and simulation tools can be developed with the robustness and reliability to handle the demands of emerging intelligence, domestic, and military technology. It will also look at limitations on modeling and simulation with respect to these increasingly complex systems.