Graph Theory Blink 10 (3 rules of geometric deep learning: locality, aggregation, and composition).

#graphNeuralNetworks #geometricDeepLearning #graphConvolutionalNetworks

Lecture 10 is a brief introduction to geometric deep learning: an exciting research field intersecting graph theory and and deep learning.

In this lecture, I cover the three fundamental rules driving the field of deep learning including:
1) Locality: “tell me who your neighbours are, I will tell you who you are”,
2) Aggregation: “how to integrate information or messages you get from your neighbour?”, and
3) Composition: “how deep you want to learn from your neighbours’ messages?”

**** Resources and further readings ****

1. Stanford course “CS224W: Machine Learning with Graphs”, offered by Jure Leskovec:
A special thanks to my students Alin Banka and Inis Buzi for sharing this with me :)
2. Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C. and Yu, P.S., 2019. A comprehensive survey on graph neural networks. arXiv preprint arXiv:1901.00596.
3. Zhou, J., Cui, G., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C. and Sun, M., 2018. Graph neural networks: A review of methods and applications. arXiv preprint arXiv:1812.08434.
4. Graph-based deep learning literature in top conferences:

[arkaung]

This is super awesome! Really love lucid visualizations with great explanations!

[Ali-neel]

If we have feature vectors for the graph arcs, how will the aggregation formula be changed?

[Ali-neel]

Thank you very much for that great explanation. Could you please share simple source code for a GNN? I know I can find the python codes on Github but for me, they are really complicated.