9/12/2025 Bruce Adams
Meta is supporting research conducted by CS assistant professor Han Zhao into multi-domain and multi-task learning from diverse sources.
Written by Bruce Adams
With the increasing number of AI models, trustworthy machine learning must, by necessity, include the accurate transfer of shared data across domains, as well as the optimal performance of tasks assigned to it.
Meta has sponsored academic research by assistant computer science professor Han Zhao from The Grainger College of Engineering Siebel School of Computing and Data Science at the University of Illinois Urbana-Champaign.
As Zhao says, the team is working on “data with common distributions to more than one domain.” The aim is to develop novel techniques to identify and integrate data from multiple models and domains into an open-source model. The team’s goal is to learn shared representations that do not vary during shifts from domain to domain.
The second line of inquiry, key to an open-source model, is multi-task learning. As the model is developed, Zhao confirms that it is multiple source domain adaptation, or the adaptive and dynamic weighting of tasks as the model accumulates data and takes on tasks.
The agreement extends from December 2024 to December 2025. Zhao’s research group will include multiple Meta researchers who will join the regular meeting for collaborative work. The project will also involve three PhD students from Zhao’s Illinois research group. The team will submit publications to ML conferences and journals, and the code created by the research group will be publicly released to enable open source adoption of their proposed techniques.
Grainger Engineering Affiliations
Han Zhao is an Illinois Grainger Engineering assistant professor of computer science and is affiliated with electrical and computer engineering.