My primary research interest is large foundation models and generative AI, which exploits large neural network models pre-trained on massive amount of unlabeled datasets, then adapted to a wide variety of downstream tasks. To make progress, I examine this with the point of view provided by the twin windows of statistical machine learning and computer systems. Statistical machine learning provides a unified framework which combines uncertainty and logical structure to model complex, real-world phenomena, while computer systems implement the learning algorithms with the highest performance guarantees. Together they provide a powerful tool to build, deploy and risk manage large foundation models. Success on this research has the potential to revolutionarize the foundation of modern science, engineering, and business.

In the past years, my group has made contributions to data science and machine learning at a foundational level. For example, we have created a new research field named nonparametric graphical models, which integrates the power of probabilistic graphical model and nonparametric methods. This research won the IMS Tweedie award (Awarded annually by the Insitute of Mathematical Statistics to one statistician for making excellent early-career contribution to statistical theory and methodology) and the ASA Noether award (Annually awarded by the American Statistical Associations to one researcher for excellent early-career contributions to nonparametric statistics). We also developed a model-based statistical optimization theory which provides provable guarantees for nonconvex learning problems. This line of research won the Best Paper Prize in Continuous Optimization in the 5th International Conference on Continuous Optimization (Awarded every 3 years to one best paper on continuous optimization). More recently, We developed a complete theory of post-regularization inference and divide-and-conquer inference for Big Data, which serves as the basis of my NSF CAREER award. We also pushed the frontier of a new field named combinatorial inference for graphical models, which aims at developing a new uncertainty assessment theory of statistical models with nonEuclidean (e.g., graphs, partitions) parameters. My earlier research along this direction has won the Alfred P Sloan Fellowship in Mathematics. Our research also won the Best Overall Paper Award Honorable Mention in the 26th International Conference on Machine Learning, the Notable Paper Award in the 16th Interantional Conference on AI and Statistics, the Best Paper Award in the National Science Research journal.