Congratulations to Wenqian Ye, Guangtao Zheng, and Professor Aidong Zhang on receiving the Best Paper Award (Research Track) at KDD 2025, one of the world’s most prestigious conferences in knowledge discovery and the #1 conference on data mining.
Their award-winning paper, Improving Group Robustness on Spurious Correlation via Evidential Alignment, addresses the challenge of shortcut learning—when machine learning models rely on misleading correlations that reduce performance on underrepresented groups. The team’s proposed method, Evidential Alignment, leverages uncertainty estimation to detect and reduce shortcut reliance without requiring group labels, leading to improved robustness and generalization.
Ye is a Ph.D. student in computer science at the University of Virginia, advised by Aidong Zhang, the Thomas M. Linville Professor of Computer Science with joint appointments in Biomedical Engineering and the School of Data Science. Zheng completed his Ph.D. in Summer 2025, also under Zhang’s guidance.
“In this work, we study the problem of improving group robustness against spurious correlations,” Ye said. “We propose an evidential alignment framework that uses uncertainty estimation to identify and mitigate shortcut features without group annotations, and extensive experiments on real-world datasets verify its effectiveness over existing baselines.”
KDD 2025 took place in Toronto, Canada, from August 3 to 7. The Best Paper Award recognizes innovative scholarly articles that significantly advance the field of knowledge discovery and data mining. Each year, the award is given to authors of the strongest paper selected through a rigorous review process.
