Taufer Leads NSF-Funded $9 Million Team to Accelerate AI-Driven Discovery

A project led by Michela Taufer, a MathWorks Professor at the University of Tennessee, Knoxville, has received a $9 million grant from the National Science Foundation to accelerate AI-driven scientific discovery. The effort focuses on turning fragmented, hard-to-reach research data into a national capability that supports secure discovery, access, analysis, and sharing across the U.S. research ecosystem.

Taufer is coordinating a multidisciplinary team spanning UT, the University of Utah, Purdue University, the Texas Advanced Computing Center, and MLCommons, alongside partners from universities, national laboratories, and industry. The approach targets a common bottleneck in modern science: researchers increasingly depend on high-performance computing resources, but access to those resources is uneven and collaboration can be slowed by data movement and management overhead.

In today’s experiments, instruments stream enormous volumes of information every second, often in heterogeneous formats. Even when data is available, converting it into usable scientific insight typically requires substantial compute, careful organization, and long preparation cycles—sometimes taking months before analysis can begin. Additionally, collaborators may be unable to participate if they lack compatible infrastructure or permissions.

Taufer describes the goal as increasing scientific throughput while making advanced AI-enabled workflows accessible regardless of an institution’s HPC capacity. By lowering these barriers, the project aims to improve reproducibility, expand participation by educators and students, and enable teams to iterate more rapidly toward measurable discoveries.

The work builds on earlier NSF funding received in 2022 through the Integrated Data and Systems Sciences program, which supported development and pilot testing of the National Science Data Fabric. NSDF is designed as a secure digital layer that allows researchers to connect to data where it is generated—such as leadership-class computers, campus clusters, or scientific instruments—without forcing them to physically relocate data.

A recent demonstration helped illustrate NSDF’s real-time collaboration potential across regions. Scientists at Cornell’s Structural Materials Beamline in New York streamed live measurements while the system linked the stream to AI infrastructure at Oak Ridge National Laboratory. ORNL used the incoming data to generate and update an AI model of material strain, returning guidance on where to measure next as the experiment continued.

During the pilot phase, NSDF indexed more than 75 petabytes of data across 68 repositories, spanning multiple disciplines. With the new award, the project will transition from a research prototype into a production-scale resource meant to serve far more scientists and communities, moving toward an ongoing national “data-to-decision” workflow.

Scaling the concept also requires interoperability across very different facility types, each with distinct data characteristics, policies, and technologies. The team will continue investing in cyberinfrastructure, AI and data management, while expanding the community of domain scientists, engineers, and computer scientists working to integrate these systems as one coherent platform.

Keywords

Artificial intelligence; AI-driven discovery; scientific data fabric; cyberinfrastructure; secure data sharing; HPC accessibility