VALERIE TSAO
Project 5: AI-Driven Simulation Agents for Urban Infrastructure Risk Modeling (Zheng / Li)

Valerie is a 2nd-year PhD student at Duke University in Theoretical, Computational, and Applied Mechanics. Her research sits at the intersection of physics-aware machine learning and probabilistic forecasting, using generative models constrained by physical laws to capture the distribution of stochastic events acting as a foil to classical deterministic surrogates. She recently published a probabilistic interpolation framework for reconstructing sparse temperature dynamics using diffusion-based inpainting enriched with geostatistical techniques, and is currently developing a method that trains directly on sparse data by integrating a Maximum a Posteriori objective with the Advection-Diffusion PDE. She also develops generative frameworks for modeling extreme storm events over the San Jacinto river basin. Originally from Taiwan, Valerie’s drive to study climate is deeply personal shaped by witnessing typhoon damage and learning that rising sea temperatures have threatened the marine ecosystems of her home island.