Course talk:CPSC532:StaRAI2020:GraphNeuralNetworks

The information shared on this page provides a good high-level overview of GNN. But, some pictures or tables differentiating different GNN would be helpful. Also, a small table summarizing each network(learning setting or equation) would be helpful. Also, there is an incomplete section "related pages"(I think this needs to be deleted).

Thank you, Maulik for your feedback. We have added the final image of the different networks and I hope it would help.

The information shared on this page is good, however, it doesn't really cover any intricacies of the topic at hand. I'm not quite sure how the mapping occurs, what are the limitations of such mapping or how much knowledge is learnedt. Covering that can be quite helpful.

I think it would be good to have a specific example of one of the GNNs models (the most basic one) so that we can see how it works. And then explain how the other models extend this. At the moment is too superficial for your peers to be able to understand GNNs from your description. This of this as a teaching moment; you want to teach your peers about GNNs so they get the basic idea(s).

Also, what does a GNN represent / learn? Properties of nodes? Properties of arcs? Missing arcs? Properties of the whole graph? Do they assume the graph structure is known? For learning a graph, be explicit about what is input and what is output.

(Also please fix the math; sometimes you use $ when you should use <math> (I think)).

How does learning/training work on a GNN? Is there any concept of forward and backward propagation in a graph that is not a DAG?

(Also, some pictures would help.)