Harnessing the Data Revolution to Enable Predictive Multi-scale Modeling across STEM

Development of lightweight materials for cars and airplanes, prediction of the weather in space and on earth, and design of new drugs for medical treatments are examples of important societal problems that require modeling of physical processes that occur at multiple and vastly different scales. For example, the external deflection of a composite airplane wing can be measured in inches but the displacements of the molecules inside that composite material, which ultimately govern the deflection, are a fraction of the thickness of a hair.

With continuing advances in computers and computational modeling, it is now possible to reliably predict what happens at any of these scales. However, it is still challenging to design predictive models across multiple scales. This National Science Foundation Research Traineeship (NRT) project will facilitate research and training that combines novel artificial intelligence and machine learning techniques with traditional computer simulations and high-performance computing.

The new methods will preserve physical structure from the small to the large and enable the discovery of new science and engineering applications of importance to society. The project anticipates training 100 Ph.D. students, including 35 funded trainees, from a variety of areas, including Computational Mathematics, various Engineering disciplines, Mathematics, Statistics and Probability, Biochemistry and Molecular Biology, and Physics and Astronomy.

The research and training in this project are targeted at modeling multi-scale phenomena where well-formed and validated modeling hierarchies exist but where traditional modeling and simulation techniques break down. Recent years have witnessed the development of machine learning approaches and their broad impacts in computational modeling and scientific computing.

Despite the overwhelming success machine learning has had in other areas such as natural language processing and image analysis, it is unlikely that modeling exclusively based on machine learning can replace traditional simulations that explicitly incorporate the domain knowledge arising from the physical mechanisms, symmetries, and constraints. This project will advance training and research in predictive modeling of multi-scale phenomena in complex fluids, biophysics, and polymeric materials by advancing the hybridization of traditional accurate, high-performance numerical methods with modern structure-preserving machine learning techniques.

The project will develop a systematic approach to train a hierarchy of models that result in tractable simulations using high-fidelity computed and experimental data. These will unlock challenging problems in the application areas of the project and beyond. The project will also deliver a model for educating tomorrow’s graduate students and STEM workforce in predictive modeling of multi-scale phenomena.

It will include core coursework leading to a graduate certificate, a course on scientific communication, and targeted short courses in cutting-edge topics, as well as an internship at a partner site and a rigorous professional development program. Upon leaving the program, the trainees will be ready to pursue careers in research and training in academia, industry, or national labs.

The NSF Research Traineeship (NRT) Program is designed to encourage the development and implementation of bold, new potentially transformative models for STEM graduate education training. The program is dedicated to effective training of STEM graduate students in high priority interdisciplinary or convergent research areas through comprehensive traineeship models that are innovative, evidence-based, and aligned with changing workforce and research needs.

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.