Patrick Gallinari

We introduce a generative masked autoregressive neural operator that models parametric PDE dynamics in latent space, enabling efficient trajectory generation, in-context learning, and few-shot generalization from irregular data.