Han Zhao, assistant computer science professor at Siebel School of Computing and Data Science in The Grainger College of Engineering at the University of Illinois Urbana-Champaign, is the recipient of two NSF awards.
“Needless to say,” he notes, “I was thrilled to hear the good news from my NSF program directors, especially during these challenging times. I am thankful to all the unwavering support and guidance offered from the school, my mentors, and my colleagues. I want to thank all my students for their sustained effort in pushing our research agenda forward.”
He has received a prestigious NSF CAREER award for his research on “Efficient and Trustworthy Machine Learning via Post-Processing.” Zhao’s project directly addresses the weaknesses of language models – accuracy, robust generalization and interpretability – in three thrusts. The project, funded with $600,000 through July 2030, is focused on group fairness, generalization across tasks and domains, and data attribution to enhance trustworthiness. Post-processing techniques, scalable to LLMs, will be utilized, and project outcomes will be incorporated into undergraduate and graduate courses, as well as shared as open-source software packages.
Zhao is also the recipient of an NSF CISE CORE award for “Small: Neural Probabilistic Circuits: Towards Compositional and Interpretable Neuro-Symbolic AI.” The grant is for $500,000 for a period ending in August 2028. The project team will examine deep neural networks, whose black box nature makes their decisions difficult to interpret and explain. Neural Probabilistic Circuits are proposed to convert black-box predictions to “a more transparent process by integrating neural networks for pattern recognition and probabilistic circuits for tractable reasoning.”
Zhao explains, “Deep neural networks are often called ‘black boxes’ because, despite their impressive performance, it’s difficult to understand exactly how they make decisions. This is particularly the case nowadays when increasingly larger models are being trained and deployed in many real-world applications. They work by learning complex, high-dimensional functions that lack transparent, human-understandable structure.”
The team’s research will inform course material and be shared via open-source software.
Trustworthy Machine Learning Group research areas
Commenting on the similarities between the two award-winning projects, Zhao notes that “our group is called the Trustworthy Machine Learning Group, and that’s precisely our research focus. The CAREER award will focus on the robustness, fairness, and privacy aspects of modern machine learning systems. The CISE CORE award attempts to propose and build a neuro-symbolic AI system that is interpretable by design.”
Grainger Engineering Affiliations
Han Zhao is an Illinois Grainger Engineering assistant professor of computer science.