Xiaopeng Jiang, assistant professor of computer science, center, along with doctoral students, Fahad Hossain, left, and Zhicheng Yang, work in Jiang’s lab. Jiang has received a $175,000 grant from the National Science Foundation’s Computer and Information Science and Engineering Research Initiation Initiative to pursue “united learning.” (Photo by Russell Bailey)
March 18, 2025
SIU researcher gets $175K grant to make AI ecosystems more efficient
CARBONDALE, Ill. – With advances in AI often outstripping the capacity of existing hardware, a researcher at Southern Illinois University Carbondale is looking at ways to leverage the existing computing environment to keep up with demands for training AI models.
Xiaopeng Jiang, assistant professor of computer science, has received a $175,000 grant from the National Science Foundation’s Computer and Information Science and Engineering Research Initiation Initiative to pursue “united learning.” The computing approach allows users to leverage diverse computing power together, including PCs, smartphones and Internet-of-Things (IoT) devices, to train large AI models – avoiding the need for costly new processors.
Emerging technological advances in the industrial setting or in smart homes could benefit from the united learning approach, by performing such tasks as visual quality inspection, predictive maintenance and predictive air conditioning control.
“There were an estimated 1.5 billion PCs worldwide in 2024, with up to 80 to 90% of their CPU power going unused at any given time,” Jiang said. “By harnessing the idle computing power of these PCs, along with resource-constrained devices such as smartphones and Internet-of-Things devices, we can create a more efficient, accessible and sustainable AI ecosystem that benefits society as a whole.”
Jiang’s two-year grant is part of a highly competitive and prestigious NSF program aimed at supporting research independence among early-career academicians. The program involves developing research capabilities in future generations of computer and information scientists and engineers, including computational and data scientists and engineers. Jiang’s grant mostly will support two doctoral students but also will pay for conference travel.
Current distributed AI learning frameworks are often based on “federated” architecture. Federated systems involve multiple, independent devices sharing resources to pursue a common goal. Federated systems can be established among various organizations, software applications or networks, with the participants cooperating on standards, protocols and policies to facilitate coordination and interoperability.
As the prevailing AI learning framework, however, federated systems-based learning models don’t allow data sharing among devices.
On the other hand, Jiang is developing so-called united learning, which unites all devices across a distributed system to train complex models efficiently while allowing data sharing among devices, thereby increasing model accuracy and eliminating the need for mass hardware upgrades for AI models, such as expensive graphics processing unit (GPU) servers.
An example might be a hospital using its existing computers to train a large language model in health care services. It could do so without sending any sensitive patient information to external providers, while sharing internally.
“It can help health care and educational institutions leverage distributed computing resources to train large language models tailored to their needs,” Jiang said.
Jiang’s doctoral students, Fahad Hossain and Zhicheng Yang, will assist him with this project. Jiang also plans to develop new learning modules on data mining and big data analysis, as well as design an undergraduate workshop on electronics.
(Note to editors: Xiaopeng Jiang’s name is pronounced “Shiow-PENG GEE-ang.”)