Events

CDS Colloquium: Geometry guides generalization in zero-shot learning of dynamical systems

Speaker: William Gilpin

Location: 60 Fifth Avenue, Room 7th Floor Open Space
Videoconference link: https://nyu.zoom.us/j/92408704679

Date: Wednesday, April 22, 2026

What limits our ability to model multiscale systems, like turbulent flows or biological dynamics? Classical scientific machine learning uses inductive biases to encode domain knowledge into data-driven models. Yet recent efforts to build scientific foundation models suggest that scale alone unlocks surprising generalization capabilities. I will describe my group’s efforts to understand generalization in scientific machine learning. We pretrain a foundation model on hundreds of thousands of dynamical systems, and discover that it develops the ability to zero-shot forecast unseen dynamical systems without retraining. Using mechanistic interpretability probes, we find that generalization capability in large models is enabled by a complex set of internal mechanisms, including zero-shot transfer across scales, in-context learning of transfer operators, and a neural scaling law relating performance to the diversity of dynamical systems encountered during training. We find that the representations learned by our model prove informative for diverse downstream biological tasks, like representing gene expression dynamics or behavioral recordings. Our work shows the potential of large-scale learning models to enable new ways of characterizing and forecasting complex dynamics.

Speaker Bio:

William Gilpin is an assistant professor of physics at UT Austin, affiliated with the Oden Institute for Computational Science and Engineering. His group studies computational nonlinear dynamics, with particular applications to systems biology and fluid dynamics. After completing his undergraduate degree in physics at Princeton, he received his Ph.D. in applied physics from Stanford, followed by a postdoctoral fellowship at Harvard’s quantitative biology initiative. His work is currently supported by an NSF CAREER award, a Cottrell Scholarship, and the Chan-Zuckerberg Initiative.