Graph Neural Ordinary Differential Equations

Extension of the graph neural network (GNN) framework to continuous time. Graph neural ordinary differential equations (GDEs) are introduced as the counterpart to GNNs where the input-output relationship is determined by a continuum of GNN layers. The GDE framework is compatible with the majority of commonly used GNN models with minimal modification to the original formulations. GDEs are shown to be effective even in cases where the data is not generated by continuous time processes.

paper: arXiv link

This repository is a work in progress. It contains examples of Graph neural ordinary Differential Equations (GDE) applied to different tasks. Several documented tutorial notebooks will be included: these notebooks are designed to be clear and useful to practicioners and researchers at all levels of familiarity with graph neural networks (GNNs) and neural ordinary differential equations. The notebooks contain abundant amounts of comments and all runnable top-to-bottom.

GDEs rely on dgl and torchdiffeq.

• 0_1_node_classification_gde: added
• 0_2_multiagent_forecasting: work in progress
• 0_3_traffic_forecasting: work in progress
• 0_4_medical_dataset: work in progress

To suggest/request additional applications or GDE models, raise an Issue or contact me via email. If you find our work useful, consider citing us:

@article{poli2019graph,
  title={Graph Neural Ordinary Differential Equations},
  author={Poli, Michael and Massaroli, Stefano and Park, Junyoung and Yamashita, Atsushi and Asama, Hajime and Park, Jinkyoo},
  journal={arXiv preprint arXiv:1911.07532},
  year={2019}
}