Scientists and engineers are constantly on the hunt for novel materials and structures that combine valuable properties in new ways.
What if they could harness the power of data science to discover fundamentally novel materials — for everything from carbon dioxide separation to better fuel cells to earthquake-proof buildings — at a far more accelerated pace?
Washington University in St. Louis' Roman Garnett, associate professor, and Alvitta Ottley, assistant professor, both of computer science and engineering at the McKelvey School of Engineering, are taking on this challenge as part of the Institute for Data-Driven Dynamical Design (ID4), led by the Colorado School of Mines with $15 million in funding from the National Science Foundation.
This interdisciplinary, multi-institutional project aims to create new theoretically grounded and experimentally validated approaches and tools to design and discover dynamical materials and structures while solving long-standing scientific challenges in the dynamical response of materials.
ID4 will address three core data science needs: new representations and learning architectures that capture the time evolution of complex materials; efficient exploration of time-dependent design spaces; and new visual analytics tools to incorporate human feedback into the design process.
"Although machine learning has already demonstrated remarkable success in automating scientific discovery, one factor that is often ignored is the deep scientific knowledge and intuition that the users of such systems have," Garnett said. "We want to capture that knowledge and intuition in the experimental design loop and make it an integral part of the process."
The process shouldn't be one-way, Ottley said. It should benefit both the algorithms and the scientists.
"The learning will happen in both directions," she said. "The machine-learning algorithms learn from the scientists, but we will develop an interactive visualization platform that will enable scientists to learn from the ML and refine their experiments."
The Mines-led institute brings together data scientists, engineers, physicists, chemists and material scientists from 11 institutions across five states: Drexel University; Harvard University; Northeastern University; Northwestern University; Princeton University; Tufts University; University of Central Florida; University of Illinois at Urbana-Champaign; Washington University; and industry partner Kebotix Inc.
“Over the last decade, we’ve seen major advancements in using big data to predict new materials. What makes this institute different is a focus on the discovery of new pathways and mechanisms rather than just targeting a final property,” said Eric Toberer, director of the Materials Science Program at Colorado School of Mines and lead investigator of the institute.
Among the members of the interdisciplinary research team are experts in machine learning, knowledge structures and visualization, as well as organic chemistry and catalysis, ion and gas transport, metamaterials and structural materials. The team includes:
- Ryan P. Adams, professor of computer science, Princeton University
- Sigrid Adriaenssens, associate professor of civil and environmental engineering, Princeton University
- Katia Bertoldi, the William and Ami Kuan Danoff Professor of Applied Mechanics, Harvard University
- Remco Chang, associate professor of computer science, Tufts University
- Adji Bousso Dieng, assistant professor of computer science, Princeton University
- Abigail Doyle, the Saul Winstein Chair of Organic Chemistry, University of California, Los Angeles
- Elif Ertekin, associate professor of mechanical science and engineering and Andersen Faculty Scholar, University of Illinois at Urbana-Champaign
- Roman Garnett, associate professor of computer science and engineering, Washington University in St. Louis
- Diego Gomez-Gualdron, assistant professor of chemical and biological engineering, Colorado School of Mines
- Jane Greenberg, the Alice B. Kroeger Professor and director of the Metadata Research Center, Drexel University
- Sossina M. Haile, the Walter P. Murphy Professor of Materials Science and Engineering and of applied physics, Northwestern University
- Boris Kozinsky, associate professor of computational materials science, Harvard University
- Steven Lopez, assistant professor of chemistry and chemical biology, Northeastern University
- Alvitta Ottley, assistant professor of computer science and engineering, Washington University in St. Louis
- Semion K. Saikin, chief science officer, Kebotix Inc.
- Fernando Uribe-Romo, associate professor of chemistry, University of Central Florida
The institute’s work also will include a robust outreach program, including summer coding schools for high school students; undergraduate research opportunities; and one-year internships for recent college graduates who attended schools without undergraduate research programs. In the latter years of the five-year program, the institute also plans to offer graduate student fellowships, where doctoral students from universities not affiliated with the institute could come and learn the tools, contribute to their development and then bring them back to their own labs.
ID4 is one of five new Harnessing Data Revolution Institutes funded through a $75 million investment announced Sept. 28 by the National Science Foundation to enable new modes of data-driven discovery that allow fundamental questions to be asked and answered at the frontiers of science and engineering.