CAREER: Model-Free Input Screening and Sensitivity Analysis in Simulation Metamodeling

这项教师早期职业发展计划(Career)奖将通过使用大规模模拟改进建模和分析，为国家健康和繁荣做出贡献。仿真是一种广泛使用的随机系统建模方法，但当模型包含大量潜在输入参数时，对这些模型的分析在计算上和统计上都变得困难。为了提高模型的可处理性，识别重要输入参数的子集，然后使用这些输入参数设计有效的仿真实验变得重要。该项目将提供新的投入筛选和敏感性分析技术，以提高大规模、复杂系统应用中满足性能和进度要求的决策能力。这些技术有潜力应用于广泛的应用领域，如生物医学研究、医疗保健、制造以及国防和国土安全行动。该项目还将对工程教育产生积极影响，并扩大工程企业中代表性不足群体的参与。研究人员将开发方法来提高基于元模型的模拟分析的可扩展性和采样效率。其关键技术包括(1)基于Morris初等效应方法的无模型方法，用于序列输入筛选，具有严格的统计性能保证；(2)构建知识和数据驱动的可伸缩的异方差对偶元模型的理论和方法；(3)基于元模型的严格的全局灵敏度分析方法，用于量化具有有限样本和大样本性能保证的异方差下每个活动输入的影响。该项目将为有效和高效的在线投入筛选和全球敏感度分析产生新的方法和实用的算法。开发的方法和算法将通过与生物医学信息学研究人员正在进行的两项研究合作，在复杂的随机系统上进行测试。该教育计划包括为女性工程专业学生开发大学前推广计划、本科生和研究生课程，以及与职业发展相关的研讨会。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This Faculty Early Career Development Program (CAREER) award will contribute to national health and prosperity by improving modeling and analysis using large-scale simulations. Simulation is a widely used method to model stochastic systems, but analysis of these models becomes computationally and statistically difficult when the models involve a large number of potential input parameters. In order to improve model tractability, it becomes important to identify a subset of significant input parameters, and then to design effective simulation experiments using these input parameters. This project will provide new input screening and sensitivity analysis techniques for improving the decision-making capability within performance and schedule requirements in large-scale, complex systems applications. The techniques have the potential to apply to a wide range of application areas, such as biomedical studies, health care, manufacturing, and defense and homeland security operations. This project will also positively impact engineering education and broaden the participation of underrepresented groups in the engineering enterprise. The investigator will develop methods to enhance the scalability and sampling efficiency of metamodel-based simulation analysis. The key technical components include (1) a model-free method based on the Morris elementary effects method for sequential input screening with rigorous statistical performance guarantees, (2) theory and methods for constructing knowledge- and data-driven scalable heteroscedastic dual metamodels, and (3) a rigorous metamodel-based global sensitivity analysis approach to quantifying the impact of each active input under heteroscedasticity with finite-sample and large-sample performance guarantees. This project will generate novel methodology and practical algorithms for effective and efficient online input screening and global sensitivity analysis. The methods and algorithms developed will be tested on complex stochastic systems through two ongoing research collaborations with biomedical informatics researchers. The educational plan includes development of pre-college outreach programs, undergraduate and graduate curricula, and career development-related workshops for female engineering students.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.