Matteo Rufolo

In-context learning now offers calibrated, long-horizon predictions for system identification, rivaling traditional methods and even generating synthetic data for parametric models.

A probabilistic approach to manifold meta-learning achieves high predictive accuracy while providing reliable uncertainty quantification in low-data environments.

Diffusion models can dramatically improve the robustness of robot dynamics models to distribution shift, outperforming deterministic Transformers in challenging conditions.