FAENet: Frame Averaging Equivariant GNN for Materials modeling

🌟 This repository contains an implementation of the paper FAENet: Frame Averaging Equivariant GNN for Materials modeling, accepted at ICML 2023.🌟 More precisely, you will find:

  • FrameAveraging: the data transform that projects your 3D graph into a canonical space of all euclidean transformations, as defined in the paper.

  • FAENet: a GNN architecture for property prediction on 3D atomic systems.

  • model_forward(): a high-level forward function that computes appropriate equivariant model predictions for the Frame Averaging method, i.e. handling the different frames and mapping to equivariant predictions.

More information are provided in the User Guide.

See the source code for the package on `Github <https://github.com/vict0rsch/faenet>`_

Contact

Alexandre Duval (alexandre.duval@mila.quebec) and Victor Schmidt (schmidtv@mila.quebec). We welcome your questions and feedback via email or GitHub Issues.