Bringing AI Education to Every College and Discipline

Universities play an important role in preparing the next generation of graduates to work with technology. However, constrained budgets and corporate competition for AI-focused faculty and researchers have left many universities falling behind. 

University-based AI knowledge and talent cultivation have often been left in a dichotomy of “haves” and “have-nots.” 

With limited access to AI hardware, data, and software and few AI-trained professors, many students are graduating without the skills they need to succeed in an increasingly AI-driven world. Public universities that lack the funding of well-resourced private universities are especially at risk.

With a goal of creating equal opportunity for students to prepare for and participate in the AI-driven future, the UF AI steering committee invited each of the 16 colleges to jointly plan and fund AI initiatives across the university.  

To get started, the university needed to raise funds to support the creation of new AI computing infrastructure, software, and laboratories and recruit new talent. 

The state of Florida provided $100 million in funding for UF to build a new data center and $15 million per year to hire 100 new AI faculty to support sustained AI education.

The steering committee also turned to its donor network to raise additional funds. 

UF spent $15 million to update their data center capacity.  

With preparations ready, NVIDIA solution architects and product engineers worked on site to install HiPerGator AI, an NVIDIA DGX SuperPOD™ consisting of 140 NVIDIA DGX™ A100 systems, each with eight NVIDIA A100 Tensor Core GPUs, and an additional 17,920 CPUs. Completed in 90 days, the installation became the largest AI supercomputer installed at a university.   

Using their DGX SuperPOD, UF developed the largest clinical language model to date, GatorTron, with 3.9 billion parameters. More than 10X larger than the second-largest clinical model, it reduced misclassification by 24 percent from BERT-large in the i2b2-2010 named-entity recognition benchmark.

HiPerGator AI has been in production since January 2021 and is already helping UF spread the benefits of AI education, conduct impactful research, and better prepare students for the jobs in demand today.

To encourage AI engagement beyond traditional STEM disciplines, UF created an AI Research Catalyst Fund to help teams pursue imaginative applications of AI. In the first call for proposals, 133 requests for funding were submitted. Twenty were selected as having the greatest potential to elevate the university’s research profile. Funded projects included: 

• AI to detect biomarkers for Alzheimer’s disease
• Machine learning to track past and present land use and patterns in Florida to measure the impact of development
• AI to help teach identify academically at-risk students
• AI to identify parasitic nematodes that damage agricultural products 

The university combined existing databases and data analysis strengths with new AI computing strengths to drive these projects forward. 

By partnering with all 12 public universities in the state of Florida university system, UF is helping to increase access to AI resources, including to groups historically underrepresented in science and engineering.

UF is undertaking moonshot research initiatives—complex projects intended to tackle the globe’s most pressing problems. 

One of the first examples of this is the university’s development of GatorTron, a natural language processing (NLP) model that enables computers to read and interpret medical language in clinical notes that are stored in electronic health records (EHRs). By automatically extracting accurate information from these notes, researchers hope to accelerate clinical research and medical decision-making.  

UF and NVIDIA used HiPerGator AI to train the GatorTron NLP model on 82 billion words and 8.9 million parameters contained in medical records. Training was completed in seven days, beating out the previous largest model, ClinicalBERT. 

This vast trove of medical insights can be applied to numerous use cases, including applications that more efficiently match patients to medical trials, helping to speed drug discovery.  

As a follow-up project, NVIDIA and UF are developing SynGator, a neural network that generates synthetic clinical data to train healthcare AI models. SynGator can generate health records of digital cancer patients, which researchers can then use to create tools and models without risking the privacy of protected health information.

UF was the first university in the country to introduce AI curricula across disciplines, offering elective AI courses in every undergraduate major. New elective courses include: 

• Biomedical Data Science 
• AI In Agricultural and Life Sciences 
• AI In the Built Environment
• AI In Media and Society

The university offers a nine-credit AI certificate for undergraduates to acquire a baseline of AI literacy. The certificate requires students to complete an AI fundamentals course, an AI ethics course, and a major-specific course. 

To help spread AI knowledge beyond university walls, UF also launched a microcredential designed for working professionals. It provides training and digital certificates to demonstrate in-demand AI skills, know-how, and experience.