A Hitchhiker's Guide to Geometric GNNs for 3D Atomic Systems

Explore a comprehensive overview of Geometric Graph Neural Networks (GNNs) for 3D atomic systems in this 1-hour 21-minute conference talk by Simon Mathis, Chaitanya Joshi, and Alexandre Duval from Valence Labs. Delve into the fundamental concepts, including background material and a pedagogical taxonomy of Geometric GNN architectures. Learn about invariant networks, equivariant networks in Cartesian and spherical bases, and unconstrained networks. Discover key datasets, application areas, and potential future research directions in this field. Gain insights into the inductive biases leveraged by Geometric GNNs, such as physical symmetries and chemical properties, to learn informative representations of geometric graphs. Follow the modelling pipeline and explore various types of Geometric GNNs, including invariant, equivariant, and unconstrained models. Conclude with a Q&A session to further enhance your understanding of this cutting-edge topic in AI for drug discovery and computational modelling of atomic systems.

- Intro + Background
- Geometric GNNs
- Modelling Pipeline
- Invariant Geometric GNNs
- Equivariant GNNs
- Other Geometric "Types"
- Unconstrained GNNs
- Future Directions
- Q+A