GraphN

A convenient wrapper to develop graph neural networks with Keras. Currently under development with the objective of integrating Networkx, Owlready2 and oneM2M for structured deep learning models in IoT.

Papers to implement

A non-exhaustive list of relevant papers:

• Semi-Supervised Classification with Graph Convolutional Networks, Kipf, T. N., & Welling, M. (2016)
• Modeling Relational Data with Graph Convolutional Networks, Schlichtkrull, M. et al. (2018)
• Graph Attention Networks, Velickovic, P. et al. (2017)
• Hierarchical Graph Representation Learning with Differentiable Pooling, Ying, Z. et al. (2018)
• Gated Graph Sequence Neural Networks, Li, Y. et al. (2015)
• Variational Graph Auto-Encoders, Kipf, T. N., & Welling, M. (2016)

Some interesting features to implement

It would be awesome to work easily with directed labeled multi-graphs, as it's a nice representation for relational data. Being able to easily interwork with ontologies is also quite convenient. At an higher level, enabling routines to handle oneM2M's base ontology would simplify the development of deep learning applications in the IoT.