Photo above courtesy of Cornell University.
Michela Taufer, MathWorks Professor in the Tickle College of Engineering, is the principal investigator on a $9 million project awarded by the National Science Foundation to accelerate AI-driven scientific discovery by enabling researchers to securely discover, access, analyze, and share scientific data across the nation’s research infrastructure.
She is leading a collaborative team comprising UT, the University of Utah, Purdue University, the Texas Advanced Computing Center, and MLCommons, with additional collaborating institutions across academia, national laboratories, and industry.
“We are living through a scientific renaissance driven by data and AI,” Taufer said. “Modern scientific instruments generate enormous amounts of data every second.”
The sheer volume of data—and the many forms in which it comes—requires significant computing power to find, retrieve, and analyze. Researchers must have access to high-performance computing (HPC) infrastructure, and even then, they may spend months moving and organizing data before they can use it. Collaborators may still not have access.
“The challenge is turning all that data into discoveries quickly enough to accelerate scientific innovation,” Taufer said. The solution must increase process efficiency while making advanced, AI-driven science accessible to any researcher, regardless of their HPC resources.
“By lowering these barriers, we can enable every U.S. researcher, educator and student at institutions of all sizes to contribute new ideas, improve reproducibility, and foster collaboration,” Taufer said. “Our goal is to transform scientific data into scientific decisions in real time.”
Building—and Demonstrating—the National Science Data Fabric
In 2022, Taufer and partners received funding through the NSF Integrated Data and Systems Sciences program to build and pilot the National Science Data Fabric (NSDF).
Typically, researchers must find and move data from its source to their own computing infrastructure. NSDF changes that paradigm. Researchers can use this national digital infrastructure for scientific discovery to securely connect with data wherever it is generated—for example, a national laboratory’s leadership-class computer, a university campus cluster, or a specific scientific instrument.
“Imagine a student at a small university without a large computing infrastructure working with petabytes of NASA satellite data, physicists around the world collaborating through shared dark matter datasets, or scientists using AI to guide experiments at national facilities in real time and analyze data as it’s produced,” Taufer said. NSDF makes opportunities like these possible.
Earlier this summer, Taufer’s team worked with partners at the University of Utah, Cornell University, and Oak Ridge National Laboratory (ORNL) to demonstrate NSDF’s ability to transform real-time collaboration between institutions located in different regions of the country.
In New York, Cornell High Energy Synchrotron Source scientists operated the Structural Materials Beamline (SMB) to examine an additively manufactured stainless steel wall. NSDF connected the SMB with ORNL’s AI infrastructure while the experiment was running. SMB sent its real-time data; ORNL’s infrastructure used that data to create, maintain, and update an AI-driven model of the strain within the metal wall and send back recommendations about where to measure next. Researchers in three states monitored the process simultaneously.
“NSDF provides the digital backbone that connects experimental facilities, AI services, computing resources, data repositories, and scientists into a unified research ecosystem while supporting reproducible and reusable workflows,” Taufer said.
Scaling Up Success
During the pilot phase, NSDF indexed more than 75 petabytes of data across 68 repositories, demonstrating that data can be securely shared and managed across institutions and scientific disciplines.
That success, large as it is, represents a fraction of data being generated. With the new NSF award, Taufer and her colleagues will transition NSDF from a successful research prototype into a production-scale national capability serving researchers across scientific disciplines.
“NSDF is the digital foundation for a shift in how science actually gets done, away from the old sequential model,” Taufer said.
To achieve that, Taufer’s team must continue figuring out how to enable very different types of facilities to work together as one collaborative, seamless system. “A synchrotron, a neutron source, a satellite mission, and a supercomputer each generate different kinds of data using different technologies and policies,” she noted as an example.
Integrating such different technologies underscores the need to not only keep advancing cyberinfrastructure, AI, and data management, but to keep growing the NSDF community of computer scientists, engineers, domain scientists and partners engaging across more disciplines.
Tennessee at the Forefront
UT researchers have been advancing HPC infrastructure for decades. Now, by heading statewide initiatives like AI Tennessee and national efforts like NSDF, the university is helping shape the future of AI-enabled scientific discovery through partnerships spanning universities, national laboratories, industry, and scientific facilities.
UT is also training the next generation of scientists and engineers in modern cyberinfrastructure and AI-enabled workflows. Students and early-career researchers will help design, build, and deploy the technologies that drive the NSDF.
“Just as UT played a leadership role in the computing revolution, we are now leading the transition to AI-enabled scientific discovery, where data collected anywhere can become knowledge everywhere,” Taufer said. “That leadership strengthens Tennessee’s position as a national hub for advanced computing, AI, and data-driven science.”
This article originally appeared on news.utk.edu.