This is probably the one of the best hands-on learning resources I've come across for GNN! Looking forward to this project. And you got yourself a subscriber, of course! :)
Good question :) I'm not entirely sure yet. I've worked with graph VAEs before, so probably it's going to be them. I also plan to give an overview on what possible alternative methods exist.
@@DeepFindr I'm doing a research to know which is the best model I should use, so I'm testing both of them to benchmark ... for me i work with dgl ... i hope the best and good job !
Great project and video! Will you use GIN (Graph Isomorphism Networks, like MolGIN) to build the model or a Graph Transformer? GINs are described in the Stanford lecture on UA-cam: CS224W: Machine Learning with Graphs | 2021 | Lecture 9.2 - Designing the Most Powerful GNNs
Hi! Thank you! Yes I was also following the Stanford series, great content! As the layers are easily exchangeable in pytorch geometric I will probably try out different things. So far I tried Graph Attention Networks and classical GCNs. GIN and Transformer Conv are also available in PyG, I will certainly try them out! Thanks for the hint!
@@DeepFindr The MolGIN paper „Enhanced Graph Isomorphism Network for Molecular ADMET Properties Prediction“ shows how to take bond information of molecules into account. However, a good representation of cycles in molecules is currently missing in GINs. A well performing GNN that can be used as benchmark is AttentiveFP as it is quite robust. I really love your Explainable AI series! 😀
@@torstenschindler1965 thanks for the hints! I see you have deep knowledge about the literature :) I'll have a look at the resources. At the moment I search for the best way to combine the node embeddings to a graph embedding. I play around with different pooling techniques as well as dummy nodes. Let's see :)
Hi ,as you know am top fun to your videos. Am really so busy recently with some personal problems but i cant leave with out put my happiness and support your idea ,rabidly i have few comments it will be fantastic if you can do it and be as intensive course learning : 1-put the problem and introduce it in all aspects 2-state of art of some methods and view to solve it and put door to improve your solution to give us way to interact with you and may be this will creativity space. 3- mention all the pre requirements to run the experiment, google colab ,or other and what is tge alternatives if one do not have each. 4- develop story tell to formalize the report and also i request to be work sharing with all audience and am sure most want to involve in such thing to learn and build skils . Last put one session online to ask and communicate all after each step to be sure all in one way.
![The moment we stopped understanding AI [AlexNet]](/img/n.gif)
This is probably the one of the best hands-on learning resources I've come across for GNN! Looking forward to this project. And you got yourself a subscriber, of course! :)
Thank you! I appreciate it :)
I can't wait to see the incoming videos.
super excited to see a live implementation.. I am working on recommender system using GNNs
Sounds like a useful series on an interesting subject. I'll join along!
Looking forward to this series.
Great work.. excited about it
Are you going to make it for heart disease prediction
Thanks for the idea but I want to ask about what generative models do you plan to use? ( VAE or GAN ... )
Good question :) I'm not entirely sure yet. I've worked with graph VAEs before, so probably it's going to be them. I also plan to give an overview on what possible alternative methods exist.
@@DeepFindr I also working with VGAE but my goal is to apply Graph gan ! another question please will you use pyg or dgl ?
Hi,
Nice! Is there a particular reason why you switch from VAE to GAN?
This series will be with PyG - which one are u using?
@@DeepFindr I'm doing a research to know which is the best model I should use, so I'm testing both of them to benchmark ...
for me i work with dgl ... i hope the best and good job !
Nice sounds good!
I also plan to have a look at DGL soon :)
Wish you all the best as well!
You are great sir ❤️❤️
Amazing videos!!! :D
Great project and video! Will you use GIN (Graph Isomorphism Networks, like MolGIN) to build the model or a Graph Transformer?
GINs are described in the Stanford lecture on UA-cam: CS224W: Machine Learning with Graphs | 2021 | Lecture 9.2 - Designing the Most Powerful GNNs
Hi! Thank you!
Yes I was also following the Stanford series, great content!
As the layers are easily exchangeable in pytorch geometric I will probably try out different things. So far I tried Graph Attention Networks and classical GCNs.
GIN and Transformer Conv are also available in PyG, I will certainly try them out! Thanks for the hint!
@@DeepFindr The MolGIN paper „Enhanced Graph Isomorphism Network for Molecular ADMET Properties Prediction“ shows how to take bond information of molecules into account.
However, a good representation of cycles in molecules is currently missing in GINs.
A well performing GNN that can be used as benchmark is AttentiveFP as it is quite robust.
I really love your Explainable AI series! 😀
@@torstenschindler1965 thanks for the hints! I see you have deep knowledge about the literature :)
I'll have a look at the resources.
At the moment I search for the best way to combine the node embeddings to a graph embedding. I play around with different pooling techniques as well as dummy nodes. Let's see :)
is there any githu repo for this ?
Yes in the video description of the following videos :)
Very interesting
Hi ,as you know am top fun to your videos. Am really so busy recently with some personal problems but i cant leave with out put my happiness and support your idea ,rabidly i have few comments it will be fantastic if you can do it and be as intensive course learning :
1-put the problem and introduce it in all aspects
2-state of art of some methods and view to solve it and put door to improve your solution to give us way to interact with you and may be this will creativity space.
3- mention all the pre requirements to run the experiment, google colab ,or other and what is tge alternatives if one do not have each.
4- develop story tell to formalize the report and also i request to be work sharing with all audience and am sure most want to involve in such thing to learn and build skils .
Last put one session online to ask and communicate all after each step to be sure all in one way.
Hi! Thanks for your feedback :)
I wish you all the best!
can you send me the dataset pls
+