Hi all,
I’m looking for ways to learn Graph Neural Networks, I know of Jure Leskovec’s lectures on youtube
And that tensorflow has a library
What are good projects that I could do to learn?
Is there a chance of this topic being covered by DeepLearning.AI in the future?
Maybe a Specialisation in this topic would be nice from DeepLearning.AI lets hope in the future.
I think the first step is to build a few projects using a dataset similar to Jure Leskovec’s assignment. I have also done a few assignments from his course. Maybe I am thinking to build a project for 3d Generation. I hope you will try something similar for you.
I’ve been listening to this today:
Livestream of Stanford Graph Learning Workshop 2022 - YouTube 9/28/2022.
Hi Sam, I am very very interested in this topic too! We can learn together!
This is a pretty good course that comes with some assignments!
This is a GNN tutorial in a paper.
This is a repo of tutorial notebooks for GNN in molecular structures.
This is a GNN fraud detection open-sourced algorithm but probably not for the beginning.
This is a collection of some benchmarked datasets, but I think we my just use other smaller and more handy datasets.
Raymond
These are great resources!
Yeah let’s learn together!
I’ll start a project and post progress here periodically.
Sam
I am thinking about what I would want to do. Will come back with something in a week! 
Raymond
Hey @SamReiswig !
I think I am going to write a few articles of notes on GNN.
Also these articles are pretty good ones!
Raymond
Making articles is a great idea!
I asked around about resources last week and got recommended these:
They are also mentioned in the workshop livestream but I had missed that part.
I’ve also found other resources:
Zak Jost was interviewed on Machine Learning Street Talk a podcast / youtube channel on ML that I like to listen to and has some material of his own.
I’m working on making a simple submission to the OGB Benchmark.
Great stuff! I am going to try GraphSage first and I will use the OGB package.
This Graph Representation Learning Book (pre-publication pdf available) discusses some aspects of graph neural network implementation in detail and more.
Thanks Raymond,
I’m also interested in graph ML. I’ll read-around these sources you have provided!
I’m interested in ML challenges with respect to being more human understanble, instead of being just ablack box. I feel the visual apsects of graphs could lend themselves well to that.
What kind aspects of graphs interests you?
Michel
Hello Michel,
Causal graph. Causal graph interests me the most, but I am looking for a way to approach it under the subject of Graph Neural Network. In the mean time, anything that will open my eyes is interesting to me - anything because I am looking for a way anyway 
For delivering a human understandable picture, it pretty depends on who your audience is. If your audience wants to know what your model is capable of, then we can test the model against various situations (inputs) and, for example, compare the outputs with historic data. Sometimes it is just as important to deliver your findings in your audience’s language.
However, if your audience wants to know the causal structure of your inputs and the output, then I think it really is a total different story At least there is more analysis and modeling work to do than just simply building one ML model that connects inputs to the output
Thank you for your interest. If something about GNN comes across you and you want someone to discuss with, let me know. Maybe I will need to spend some time to read, to research, and to think, and maybe in the end my feedback may not be very insightful, but we can try, can’t we?
Cheers,
Raymond
Just saw this on my LinkedIn.
I think it’s too late for the first lecture but maybe signing up now we can get into the second one.
It says category theory but they will cover papers like Graph Neural Networks are Dynamic Programmers, Dudzik, Velickovic
I should probably also mention Petar Velickovic and Andrew Dudzik are some of the organizers
Looks very interesting! Thank you Sam. Just signed up and hope they will get back soon.
They have recordings of the lectures on YouTube.
An online Meetup Group I like to attend recently had two speakers talk about GNNs. Graph Neural Networks: Theory, Problem, and Approaches - YouTube
There’s two speakers and a Q and A section where if I remember correctly all the questions from the audience were answered.
Petar Velickovic was interviewed by Machine Learning Street Talk at NeuRIPS recently.
Sam, what do you think about those lectures? Did any one of them impress you a lot?
All of the lectures are great, I recommend all of them!
But if I had to pick my favorite three:
The “type checker” result from Petar’s Lecture week 2 is really cool! and they showed practical uses as well. The paper from the lecture: [2203.15544] Graph Neural Networks are Dynamic Programmers
The lecture by Pim de Haan week 4 on Natural Graph Networks topic is really interesting. The paper from the lecture: [2007.08349] Natural Graph Networks
The lecture by Taco Cohen, was on Category theory and Causal Models. I still need to spend more time on this one to fully understand it. It’s a long lecture and you can skim the slides here: https://cats.for.ai/assets/slides/2022-11-21-Causality-categories.pdf