Multi-fidelity Accelerated Global Search (MAGS)

该奖项将通过引入新的计算方法来指导现代制造设施（如生产个性化医疗的复杂生物制造过程）的设计和控制，促进科学进步，并为促进国家繁荣和健康作出贡献。可用于评估制造系统性能的高保真仿真模型计算量大，通常需要昂贵的数据收集，限制了它们的使用。另一方面，排队模型使用较少的计算时间，但准确性较差。在理解如何有效地利用高保真度和低保真度模型来设计和控制这样的系统方面存在很大的差距。该奖项支持推进大规模全局优化的理论基础和算法开发，从而能够使用不同精度和计算工作量的多个模型。这项研究将对美国生物制造业的发展产生广泛的影响，并将为多样化的学生提供部署和扩展这些工具所需的教育和培训，促进国家经济福利。该项目将为随机全局优化创建一个新的多保真加速全局搜索（MAGS）框架，该框架将智能地结合高保真和低保真模型，在解决方案质量统计分析的指导下动态分配计算工作量。目标是通过利用低保真模型来减少高保真仿真评估的数量，从而将模型近似（学习）与优化（搜索）相结合。本研究涉及MAGS的算法设计、可扩展性和有限时间随机分析。该项目将为计算复杂性提供理论基础，并有助于推导出动态和适应性强的分解方案，以实现大规模优化。个性化生物制造系统的设计和操作将用于测试和完善MAGS的整个开发过程。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This award will promote the progress of science and contribute to the advancement of national prosperity and health by introducing new computational methods to guide the design and control of modern manufacturing facilities, such as complex bio-manufacturing processes that produce personalized medical treatments. High-fidelity simulation models that can be used to evaluate the performance of a manufacturing system are computationally intensive and often require expensive data collection, limiting their use. On the other hand, queueing models use less computation time but are less accurate. There is a substantial gap in understanding how to effectively make use of both high- and low-fidelity models to design and control such systems. This award supports advancing theoretical foundations and algorithm development for large-scale global optimization enabling the use of multiple models of varying accuracy and computational effort. This research will have broad impact on the growth of bio-manufacturing in the U.S., and will provide a diverse population of students with the education and training needed to deploy and extend these tools, advancing the national economic welfare.This project will create a new Multi-fidelity Accelerated Global Search (MAGS) framework for stochastic global optimization that will intelligently combine high- and low-fidelity models to dynamically allocate computational effort guided by statistical analysis of solution quality. The goal is to reduce the number of high-fidelity simulation evaluations by taking advantage of low-fidelity models, thus integrating model approximation (learning) with optimization (search). This research addresses algorithmic design, scalability, and finite-time stochastic analysis of MAGS. The project will provide a theoretical foundation for computational complexity and aid in deriving a dynamic and adaptable decomposition scheme to realize large-scale optimization. The design and operation of an individualized bio-manufacturing system will be used to test and refine MAGS throughout its development.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.