S&DS Seminar: Talin Wu (Stanford)

Learning structured representations for accelerating scientific discovery and simulation

Abstract: Across most disciplines of science, e.g., physics, chemistry, biomedicine, materials, mechanical engineering, and energy, a most critical challenge is that their simulations and discoveries are typically slow due to the large-scale, complex and multi-scale nature of the system. In this talk, I will introduce my research that tackles this challenge by developing machine learning models with structured and efficient representations for accelerating scientific discovery and simulation. To accelerate scientific discovery, I developed neuro-symbolic methods which can distill the data into human-interpretable symbolic knowledge (governing equations and relational structures) and generalize to more complex data in inference. To accelerate large-scale scientific simulations, I developed structured representations to accelerate critical scientific simulations for fluid dynamics, plasma science, and generic partial differential equations (PDEs). For example, I developed a hybrid particle-fluid representation for simulating a large-scale laser-plasma interaction in a national lab facility that has important applications in physics, materials, and biomedical science. Our model is able to simulate millions of particles per time step, orders of magnitude faster than the classical solver, and significantly reduce long-term prediction error compared to strong deep learning baselines.

Bio: Tailin Wu is a postdoctoral scholar in the Computer Science Department at Stanford University, working with Prof. Jure Leskovec. He received his Ph.D. from MIT Physics, where his thesis focused on AI for Physics and Physics for AI. His research interests include developing machine learning methods for large-scale scientific simulations, neuro-symbolic methods for scientific discovery, and representation learning, using tools of graph neural networks, information theory, and physics. His work has been published in top machine learning conferences and leading physics journals, and featured in MIT Technology Review. He also serves as a reviewer for high-impact journals such as PNAS, Nature Communications, Nature Machine Intelligence, and Science Advances.