Zhun Deng

Hi! I am a postdoctoral researcher with Toniann Pitassi and Richard Zemel at Columbia University, and also part of Simons Collaboration on the Theory of Algorithmic Fairness. Previously, I completed my Ph.D. in the Theory of Computation group at Harvard University, advised by Cynthia Dwork. I am also fortunate to work with David Parkes, Weijie Su, and James Zou on various projects. 

My research interests mainly lie in reliable and responsible machine learning, both in theoretical foundations and applications. In particular, I develop formal frameworks to address algorithmic and societal challenges in modern data science such as risk control of foundation models, fairness, and privacy.  My tools mainly draw on distribution-free uncertainty quantification, (multi-)calibration, and reinforcement learning. 

Recent News

• 04/2024 - our paper on distribution-free risk control for large language models is accepted by ICLR 2024. 

• 03/2024 - A new paper on reconciling diverse opinions in reinforcement learning from human feedback is on arXiv now. 

• 02/2024 - I delivered a talk on practical theories in responsible machine learning at Center for Data Science, New York University. 

• 09/2023 - two papers accepted by NeurIPS 2023: distribution-free societal dispersion control (spotlight, top 3% among submissions) and uncertainty quantification in physics-informed nets. 

• 05/2023 - our paper about generalization theory for information bottleneck is accepted by ICML 2023.

• 01/2023 - two papers accepted by AISTATS 2023: reinforcement learning with stepwise fairness constraints and understanding multimodal contrastive learning and incoportate paired data.

• 01/2023 - two papers accepted by ICLR 2023: FIFA: making fairness more generalizable on imbalanced data and distribution-free quantile risk control.