Multiresolution Equivariant Graph Variational Autoencoder (MGVAE)

Paper

Published at Machine Learning: Science and Technology journal: https://iopscience.iop.org/article/10.1088/2632-2153/acc0d8

Presented at ICML 2022 (AI for Science workshop): https://arxiv.org/pdf/2106.00967.pdf

Authors

Truong Son Hy and Risi Kondor

Requirement

• Python 3.7.10
• PyTorch 1.8.0

Recommend using Conda environment for easy installation.

Experiments

• supervised_learning_molecules: Supervised learning of Multiresolution Graph Networks (MGN) for molecular properties prediction.
• citation_link_prediction: Link prediction on citation graphs by MGVAE.
• general_graph_generation: General graph generation by MGVAE.
• image_generation: Graph-based image generation by MGVAE.
• unsupervised_molecules: Unsupervised molecular representation learning by MGVAE.

For experiments on molecule generation, please visit our another repository: https://github.com/HySonLab/MGVAE