Medora opens new possibilities for research at UND

UND’s new Medora AI cluster gives researchers new computing power for AI, large datasets and complex research

In about 12 minutes, an artificial intelligence program running on UND’s new high-performance computing cluster, Medora, identified business challenges in 500 sets of notes from meetings with North Dakota business owners.

The test was an early step in a project that could help small business advisors recognize recurring problems and better anticipate what their clients need.

It also offered a glimpse of what researchers across UND can do with the University’s newest high-performance computing system.

Operated by UND’s Computational Research Center, the Medora AI Cluster combines 16 high-performance servers designed to support artificial intelligence and other data-intensive research.

UND faculty, staff and students can use the system at no cost, giving researchers across campus access to high-performance computing power.

“We’re punching above our weight with this system,” said David Apostal, advanced cyberinfrastructure research and education facilitator with the CRC.

Science done fast

Much of the research done at UND can be handled by an ordinary computer. Some of it can’t. Programs analyzing large datasets can overwhelm a laptop or desktop. And, while complex programs might run on them, the work may take months to finish.

That’s where high-performance computing can make a meaningful difference.

Apostal recalled working with a UND geology researcher, for example, whose program was estimated to take about six months to finish on a desktop computer. On Medora, it finished in just three weeks.

“Faster science is what everybody wants,” Apostal said. “Whether it’s social science or technology, engineering, whatever it is, they want to get their results quicker.”

Medora is particularly well suited to artificial intelligence research. Each of its 16 servers includes a graphics processing unit, or GPU, capable of performing many calculations at once, which is key for AI and other data-intensive research.

Using Medora to help North Dakota businesses

For the North Dakota Small Business Development Center, that computing power could help make better use of more than 20 years of information collected from businesses across the state.

Hosted at UND through the Nistler College of Business & Public Administration, the statewide program has accumulated consultants’ notes from thousands of meetings with North Dakota business owners.

Together those notes contain a wealth of information about challenges small businesses face. Finding the larger patterns within them, however, is difficult for a team of ND SBDC’s size.

“We don’t have bandwidth and capacity within our small team to really dig into all of the data that we have,” said Marnie Schuschke, associate state director of the North Dakota SBDC.

Using an artificial intelligence model on Medora, Apostal identified business “pain points” across a sample of about 500 notes in just 12 minutes. Schuschke hopes that analyzing the entire collection could help them identify the most common needs or hurdles.

Though the project is still in its early stages, Apostal says this proof-of-concept experiment shows Medora’s utility for organizations and researchers with large volumes of information that may be too time consuming to comb through manually.

A resource for everyone at UND

ND SBDC’s project is only one way researchers are putting Medora to use.

The system is also supporting AI workloads connected to UND’s Arctic Knowledge-Based System, an artificial intelligence research project using large datasets to improve forecasting and decision-making in extreme cold-weather environments.

But, Apostal noted, high performance computing isn’t limited to STEM research. He said social science and humanities researchers could also benefit from Medora if their work involves large collections of text, data or other information.

For example, Apostal recently met with another researcher preparing a grant proposal to use AI to analyze classroom recordings and provide feedback on the kinds of questions teachers ask.

“It’s really encouraging to see how much creativity there is out there. People are using these systems in amazing ways,” Apostal said of the variety of research he sees at CRC.

Medora joins Talon, the CRC’s other general-purpose high-performance computing system, as one of two major systems operated by the center.

After a student or faculty member requests an account, CRC staff help them determine which system fits their work and learn how to use it.

Now that Medora is up and running, Apostal hopes more people across campus will bring their research to the CRC.

“I just want as many people as possible to discover it and use it for their computational research,” Apostal said.

UND faculty, staff and students can request an account or ask questions through the CRC’s support page.