Abstract: Accurately modeling dynamic systems requires understanding the full distribution of outcomes, especially when accounting for uncertainties and biases. This talk explores the importance of distributional learning in dynamic systems, focusing on a recent project that uses Kernelized Stein Discrepancy (KSD) for post hoc calibration. Our approach leverages prior knowledge of the marginal distribution to correct prediction biases resulting from distribution shifts and sampling errors, enhancing model robustness. Through numerical experiments with climate models, we demonstrate the effectiveness of KSD-based calibration in refining predictions and improving adaptability to unseen data in complex, chaotic systems.

Bio: Tian Zheng is currently a Professor and Department Chair of Statistics at Columbia University. Her research focuses on statistics and machine learning methodology development for exploring and understanding patterns in complex data from different application domains such as biology, psychology, climate modeling, etc. Her current projects are in the fields of statistical machine learning, spatiotemporal modeling, and social network analysis. Professor Zheng’s research has been recognized by the 2008 Outstanding Statistical Application Award from the American Statistical Association (ASA), the Mitchell Prize from ISBA, and a Google research award. She became a Fellow of the American Statistical Association in 2014, a Fellow of the Institute of Mathematical Statistics in 2022, and a Fellow of the American Association for the Advancement of Science in 2024. From 2017-2020, she was associate director for education at Columbia Data Science Institute. Professor Zheng is the recipient of the 2017 Columbia Presidential Award for Outstanding Teaching. In 2021, she was recognized with a Lenfest Distinguished Columbia Faculty Award that recognizes the excellence of faculty as teachers and mentors of both undergraduate and graduate students.

Format: Hybrid

Virtual: Zoom link provided upon registration

Time: 12:00-1:30 pm EST

Free!

Registration