Speaker: Peng Chen, Georgia Tech
Title: Neural Surrogates for Fast and Scalable Bayesian Optimal Experimental Design
Abstract: This talk addresses the solution of large-scale Bayesian optimal experimental design (OED) problems governed by partial differential equations (PDEs) with infinite-dimensional parameter fields. The OED problem seeks to find optimal design of experiments that maximize the optimality criteria such as A-optimality, D-optimality, and expected information gain (EIG) in the solution of the underlying Bayesian inverse problem. These criteria are computationally prohibitive for large-scale PDE-constrained OED problems. We present fast and scalable methods to solve such problems based on Laplace and low-rank approximations, neural surrogates including derivative-informed neural operators (DINO) and latent attention neural operators (LANO). These methods exploit the geometry, smoothness, intrinsic low-dimensionality, latent causality of the parameter-to-observable map and use only a small number of PDE solves independent of the parameter dimensions to achieve high accuracy and efficiency.
Bio: Dr. Chen is an Assistant Professor at the School of CSE at Georgia Tech. Previously he was a Research Scientist at UT Austin, a Postdoc and Lecturer at ETH Zurich, and obtained his PhD at EPFL. Dr. Chen’s research is in the multidisciplinary fields of computational mathematics, data science, scientific machine learning, and parallel computing with various applications in materials, energy, health, and natural hazard. His research focuses on developing fast, scalable, and parallel computational methods for integrating data and models under high-dimensional uncertainty to make (1) statistical model learning via Bayesian inference, (2) reliable system prediction with uncertainty quantification, (3) efficient data acquisition through optimal experimental design, and (4) robust control and design by stochastic optimization.
In person attendance at this seminar is only open to Columbia Univesity affiliates. External guests are welcome to attend remotely. Please contact [email protected] if you need the Zoom link for this seminar.
