Upcoming Event: Oden Institute Seminar

From Chance to Choice: Stochastic Simulation, Learning, and Design

Xi Deng, Postdoctoral Scholar Research Associate, California Institute of Technology

Abstract

Stochastic simulation has long been a powerful tool for modeling complex physical systems, from light transport to particle transport and other stochastic PDEs. In this talk, I will present a line of work exploring how stochastic simulation can be extended beyond forward prediction toward differentiation and inverse design.

I will first discuss differentiable neutron transport, where gradients through Monte Carlo simulation enable the optimization of physical and geometric parameters. I will then introduce the Particle-Transport Neural Operator (PTNO), which learns transport solutions from noisy, low-cost stochastic simulations. With these two differentiable simulators, I will show how they perform on the same inverse shape design problem, highlighting their complementary strengths for simulation-driven optimization.

More broadly, I will discuss how stochastic simulation data can be used to train neural operators across different physical applications, ranging from fluid dynamics and heat diffusion to path-planning problems in complex geometries.

Biography

Xi Deng is a postdoctoral researcher at Caltech, working with prof Anima Anandkumar on training neural operators using stochastic pde solver with application in multiple physics processes in fusion reactors. She received her PhD from Cornell University in computer graphics. Her doctoral work involved measurement systems, light transport, and stochastic algorithms. Her research has been supported by the National Science Foundation (NSF), the U.S. Department of Energy (DOE), Cornell CIDA, Schmidt Sciences, and the Margot and Tom Pritzker Foundation. She is a recipient of the Booking.com Fellowship and Adobe Fellowship, and was an NVIDIA Fellowship finalist.