Machine Learning Seminar Series Spring 2026 | Scientific Machine Learning and Uncertainty Quantification for Digital Twins of Physical Systems

Abstract: Digital twins are virtual counterparts of complex physical systems that adaptively refine their predictive models through sensor-driven data assimilation. Once calibrated, these twins can support real-time forecasting and decision-making, including the design of data acquisition strategies via sequential optimal experimental design. However, deploying digital twins for large-scale systems remains computationally challenging due to the high cost of high-fidelity simulations, the curse of dimensionality in uncertainty quantification, and sparse, noisy observations. In this talk, I will present fast, scalable methods to address these bottlenecks, including derivative-informed operator learning, latent neural dynamics, and latent diffusion for posterior sampling. I will illustrate how these approaches enable robust, uncertainty-aware digital twins in applications such as flood prediction, tumor growth monitoring, and contaminant source tracking.

Bio: Dr. Chen is an Assistant Professor in the School of Computational Science and Engineering (CSE) at Georgia Tech. Before joining Georgia Tech in 2022, he was a Research Scientist at The University of Texas at Austin and a Postdoctoral Researcher and Lecturer at ETH Zurich. He earned his PhD in Mathematics from EPFL. His research spans computational mathematics and scientific machine learning, with an emphasis on fast and scalable methods for integrating data with physical models under high-dimensional uncertainty. He develops algorithms for Bayesian inference and data assimilation, optimal experimental design, and PDE-constrained stochastic optimization, with applications in materials, energy, health, and natural hazards. He has published in machine learning venues including NeurIPS and ICLR and computational math/physics journals such as SISC, SIOPT, CMAME, and JCP. He serves on the editorial board of Numerical Methods for Partial Differential Equations.

For more information, or for CODA guest access, please contact shatcher8@gatech.edu at least 2 business days prior to the event.