Generative AI

Probing and steering what video generation models internally represent about physics.

Learning sparse, physically structured, and uncertainty-aware dynamics in latent spaces.

Do video diffusion models know physics? Probing their activations for velocity, momentum, and gravity — and steering them.

Generative models are transforming science and engineering by enabling efficient synthetization and exploration of new scenarios for …

VENI, VINDy, VICI is an interpretable, data-driven framework for building generative reduced-order models with uncertainty quantification, combining variational autoencoders with a probabilistic extension of SINDy.