Zihan Liang 梁梓涵 | zihan.liang@emory.edu | Department of Data and Decision Sciences, Emory University

At Emory University, I study Applied Mathematics and Statistics under the mentorship of Prof. Ruoxuan Xiong.

My research broadly focuses on building reliable machine learning systems under uncertainty, particularly in settings with bias, incomplete data, or distributional shifts. I am interested in methodological questions around uncertainty quantification, robustness, and missing data, and I often draw on causal and statistical perspectives when they are useful for understanding model behavior.

I have worked on a range of applied problems—including multimodal healthcare modeling, applied NLP, and sequential decision-making—which has allowed me to study these reliability challenges across different domains and modeling paradigms.

What I'm Doing

My research focuses on building reliable and robust machine learning systems under real-world uncertainty, particularly in settings with incomplete, biased, and heterogeneous data.

I am interested in methodological questions around missing data, distribution shift, and multimodal learning, and how these factors affect model reliability and downstream decision-making. To address these challenges, I draw on tools from causal inference, representation learning, and modern deep learning, with a particular emphasis on healthcare and other high-stakes domains.

• Multimodal Learning under Missingness: developing models that explicitly account for missing-not-at-random (MNAR) patterns and treat missingness as informative signals in clinical, recommendation, and multimodal settings.

• Robust Generalization and Domain Adaptation: designing adaptive and calibrated learning frameworks that improve cross-dataset generalization under domain shift, especially in medical imaging and spatiotemporal systems.

• Causal and Reliability-Aware Representation Learning: integrating causal perspectives with deep models to improve interpretability, generalization, and bias correction in real-world data.

• Applied NLP under Noise and Imbalance: building transformer-based systems for clinical text understanding, sentiment analysis, and health event detection in noisy and imbalanced environments.