Derivative-Informed Neural Operators for Solving PDE-Governed Optimization Problems
Inside Livermore Lab via YouTube
Overview
Explore a cutting-edge machine learning framework for solving optimization problems governed by large-scale partial differential equations (PDEs) with high-dimensional random parameters in this 56-minute talk. Delve into the challenges of computationally expensive problems in Bayesian inverse problems, optimal experimental design, and stochastic optimization for optimal control. Discover a novel class of derivative-informed neural operators that accurately approximate both the mapping from inputs to PDE state and its derivatives, utilizes a scalable reduced basis architecture, and requires limited training data for high accuracy. Learn from Peng Chen, an assistant professor at Georgia Tech's School of Computational Science and Engineering, as he presents his research on scientific machine learning, uncertainty quantification, and stochastic optimization for complex systems under uncertainty.
Syllabus
DDPS | Derivative-informed neural operators by Peng Chen
Taught by
Inside Livermore Lab