grants

UNRI develops temporally consistent diffusion models for Neural Relational Inference: recovering hidden and evolving network structure from observed time-series dynamics. Beyond a single high-performance model, the project builds a unifying conceptual framework for the NRI field and an open-source library for the community, with applications ranging from brain networks to climate systems.

Hardware grant supporting a physics-informed machine learning framework for slope-scale (~50 m) Arctic weather simulations. The project combines hierarchical spatiotemporal GNNs with Neural ODEs and topography-aware physical constraints to downscale NWP forecasts for high-impact Arctic services, with avalanche forecasting as a primary pilot use case.

Awarded 4× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition.

Hardware grant for developing property-oriented generative models for molecules with hierarchical GNNs tailored to heterophilic molecular graphs, and Physics Informed Neural Networks to validate thermophysical properties. The goal is a navigable, property-aware latent space for conditional molecular design.

Awarded 2× NVIDIA RTX 6000 Ada Generation.