Solving the primary data challenge of the AI era As Oregon State University prepares to launch its Nvidia supercomputer next year, Xin Liang is developing tools to help researchers manage the flood of information it will produce.

When Oregon State University's Nvidia supercomputer comes online in the Jen-Hsun Huang and Lori Mills Huang Collaborative Innovation Complex next year, researchers will gain access to unprecedented computing power. They will be able to train larger AI models, analyze bigger datasets, and run more sophisticated simulations than ever before.

But the system's greatest challenge may not be computation. It may be data.

For Xin Liang, associate professor of computer science, developing open-source software technologies to help researchers manage and use that data efficiently has become a central focus of his work.

"We're entering an era where generating data is becoming easier and easier," Liang said. "The question is how we store it, move it, and use it effectively."

Liang joined Oregon State in January 2026 after positions at Oak Ridge National Laboratory, Missouri University of Science and Technology, and the University of Kentucky. He arrived with a National Science Foundation Faculty Early Career Development Program (CAREER) Award and a research vision focused on helping scientists navigate the growing gap between computational power and data management.

As OSU prepares for a new chapter in AI and high-performance computing, Liang, a member of OSU’s high-performance computing group, HiPCastor, sees an opportunity to ensure the university's researchers can fully benefit from the powerful systems coming their way.

A JPEG for scientific data with guaranteed error control

The centerpiece of Liang's CAREER Award project is a concept known as data polymorphism.

Though highly technical, the idea can be understood through a simple analogy.

When someone views a photo on a smartphone, they usually aren't looking at the highest-resolution version of the image. A compressed version is often sufficient for browsing, sharing, and understanding what the picture contains.

Liang wants scientific data systems to work similarly.

"Not every user needs the highest-fidelity representation of the data all the time," he said. "Different users and applications have different requirements. The size of data retrieved will not make a difference for scientific conclusions as long as these requirements are met, but the data transfer time differs."

Today, scientists often retrieve an entire dataset even if they need only a broad overview. Under Liang's approach, a system could provide multiple representations of the same data, from lightweight approximations to highly detailed versions.

He points to hurricane and climate simulations as a clear example. Tracking a cyclone over time doesn't require full-resolution data at every moment — only when a storm is actually forming.

"We may only need a sketch of the data to understand there's no feature of interest," Liang said. "If that is the case, we skip to the next time step without retrieving the full data for the current time step. When we notice the cyclone is forming, that's something we're really interested in — then we retrieve more data."

The ultimate result will be software, built for both CPUs and GPUs, that his team releases as open source, alongside a related NSF-funded effort to turn research prototypes into production-ready tools.

"The key idea is delivering the right data at the right resolution with guaranteed fidelity for a particular task," Liang said.

Powering the next wave of discovery

Liang sees opportunities to combine AI with data-management systems so computers can make smarter decisions about how information should be represented, moved, and analyzed. Those capabilities could become increasingly important as Oregon State expands its investments in advanced computing.

The university's Nvidia supercomputer is expected to support researchers from engineering, agriculture, climate science, materials science, and many other disciplines. Liang hopes the tools his group develops will help those scientists spend less time managing datasets and more time making discoveries.

"High-performance computing is becoming important for almost every field," Liang said. "When we make data easier to use, we can accelerate research across many different disciplines."