UF accelerates AI research discoveries with HiPerGator computing
The University of Florida has selected 20 research projects in artificial intelligence for accelerated development using the university’s HiPerGator supercomputer.
The program is a collaboration between UF Research, UF Information Technology and NVIDIA, which together are providing $2.4 million in seed funding and in-kind computer time and consulting for students, post docs and faculty working on specific projects.
“This program is to help faculty who need access to hundreds of GPUs to perform analysis on billions of pieces of data,” said David Norton, UF’s vice president for research. “These projects are expected to open up whole new avenues of artificial intelligence research and position the researchers to pursue other large external grants.”
The HiPerGator computer runs on high-performance graphical processing units, or GPUs, provided to the university by NVIDIA. The state-of-the-art processors are essential to these kinds of data- and processing-intensive projects.
For example, a team at UF Health is using the supercomputer to create GatorMedImage, the largest medical imaging model of its kind. More than 10 million anonymized medical images from UF Health will be used to develop and train an AI system that helps physicians to better detect and diagnose conditions and support prognoses in various organs. The system will be built on X-ray, ultrasound, CT and other images used in radiology, cardiology, obstetrics and ophthalmology.
For patients who suffer strokes, brain injuries or other neurological issues, timely and accurate diagnosis is essential but challenging. Benjamin Shickel, an assistant professor of quantitative health, is using HiPerGator to develop foundational models of patient-specific brain health by collecting and analyzing data before and after neurological events. The algorithm Shickel’s team is working on will use massive datasets and state-of-the-art generative AI to trace the evolution of a patient’s neurological changes and help doctors accurately forecast future treatment outcomes.
“Neurological diseases affect more than 43% of the global population and are the leading cause of disability and the second leading cause of death worldwide,” Shickel said. “The complexity and uncertainty of neurological conditions are direct barriers to the accurate and rapid diagnosis and clinical prognostication that are required for optimizing patient outcomes.
Mingjie Liu, an assistant professor in the Department of Chemistry, is using the computer to "harness AI to revolutionize the discovery of new molecules, making it as simple as creating AI-generated art.”
“To achieve this, we are developing an advanced AI model that learns from all known molecular structures and is fine-tuned for specific applications,” Liu said. “Using this approach, we aim to design molecules with unique optical properties for technologies such as solar cells, glowing materials and quantum devices. We believe this method will pave the way for faster and more efficient molecular design."
Over the past 12 months, HiPerGator processed 33.7 million research requests from nearly 7,000 UF researchers. HiPerGator also was used as a teaching tool in 76 courses with 2,400 students.
Kaleb Smith, director of the NVIDIA AI Tech Center at the University of Florida, works with research teams in collaboration with UFIT and NVIDIA experts to help “understand the problem they are trying to solve.”
“We brainstorm with the researcher and team to figure out what some of the best solutions might be in order to achieve their project's objective,” Smith said. “Once a plan is in place, we work with the researcher and team to help accelerate the learning curve on the NVIDIA platform.”
Smith said this could be training a lab in AI software development or working with a single PhD student running through code to identify where significant advances can be made.
“We always make sure a solution involves significant training,” he added, “so the whole team learns how to implement NVIDIA solutions, rather than just producing a single publication or project breakthrough.”