Mini-symposium (MS)

MS13: Surrogate modeling for high-dimensional problems and applications

Organizers: Hoang Tran, Oak Ridge National Laboratory

                     Armenak Petrosyan, Georgia Tech

Abstract: The approximations of high-dimensional systems from data play a pivotal role in a wide variety of mathematical and scientific problems including uncertainty quantification, control and optimization, statistical inference and data processing. Such problems often require repetitive, expensive measurements (for instance, ensemble of complex numerical simulations or time-consuming physical experiments), thus, it would be very beneficial to have access to an accurate surrogate model, which can be used in place of the original model, to approximate the input-output relationship of interest. This mini-symposium aims at bringing together people working on the theory and methods for surrogate modeling and high-dimensional approximation, in particular, but not restricted to, nonlinear approximation, sparse recovery, low-rank approximation and deep neural networks, showcasing the latest results on both methodology and applications.

Saturday, September 18, 3:30 PM – 5:30 PM: Part I of II

Room: Mell 3520

3:30 – 4:00 Sung Ha Kang, Georgia Tech, Robust Identification of differential equations from noisy data

4:00 – 4:30 Yeonjong Shin, Brown University, Plateau phenomenon in gadient descent training of ReLU networks: explanation, quantification, and avoidance

4:30 – 5:00 Keaton Hamm, University of Texas at Arlington, Optimal transport methods in nonlinear dimensionality reduction

5:00 – 5:30 Yiming Xu, University of Utah, A bandit-learning approach to multifidelity approximation

Sunday, September 19, 10:30 AM – 12:30 PM: Part II of II

Room: Mell 3520

10:30 – 11:00 Anh Tran, Sandia National Laboratory, Gaussian process and Bayesian optimization -- Bridging the gap between theory and practice in materials science

11:00 – 11:30 Oleksandr Vlasiuk, Vanderbilt University, Optimizing short-range interactions for point cloud generation

11:30 – 12:00 Vahan Huroyan, University of Arizona, Non-convex analysis of matching and embedding of image keypoints

12:00 – 12:30 Nick Dexter, Simon Fraiser University, Learning high-dimensional Hilbert-valued functions with deep neural networks from limited data