Yao Zhang

I am currently a postdoctoral researcher in the Department of Statistics at Stanford University, advised by Prof. Emmanuel Candès. I received my Ph.D. from the Department of Mathematics at the University of Cambridge, supervised by Prof. Mihaela van der Schaar. I also collaborated with Prof. Qingyuan Zhao on several projects related to causal inference. Prior to my Ph.D., I worked with Dr. Alpha Lee on lithium-ion battery diagnosis and low-data drug discovery.

Over the years, my research has spanned statistics, machine learning, and their scientific applications. In statistics, I focus on developing assumption-lean methods to analyze complex data and models. These methods can maintain statistical validity while making weaker assumptions about the data and models. Moreover, my work seeks to improve statistical power by leveraging the unique structures within each problem. Please see my related research in three different areas:

• Uncertainty quantification for black-box models (link).
• Conditional randomization tests for complex experiments (link).
• Optimization-based sensitivity analysis for unmeasured confounding (link).

In machine learning, I develop statistical methods to improve the performance of black-box models in prediction tasks and beyond. These methods cover model selection, learning, regularization and calibration:

• Step-wise selection for time-series models (link).
• Noise contrastive learning for data embeddings (link).
• Data-adaptive regularization for causal inference (link1, link2).
• Level-adaptive calibration for conformal prediction sets (link).

In addition to my primary research areas, I enjoy exploring basic sciences and their application of statistics and machine learning. You can find my interdisciplinary work in physics and chemistry (link1, link2, link3, link4).