Graph Neural Networks: A gentle introduction

❤️ Support the channel ❤️
/ @aladdinpersson
Resources that was very useful for me when learning about GNNs that you can check out for more information and from which I've used in the slides:
Cs224w: • Stanford CS224W: Machi...
distill.pub/2021/gnn-intro/
distill.pub/2021/understandin...
• Graph Neural Networks
• Intro to graph neural ...
• Theoretical Foundation...
• ICLR 2021 Keynote - "G...
Paid Courses I recommend for learning (affiliate links, no extra cost for you):
⭐ Machine Learning Specialization bit.ly/3hjTBBt
⭐ Deep Learning Specialization bit.ly/3YcUkoI
📘 MLOps Specialization bit.ly/3wibaWy
📘 GAN Specialization bit.ly/3FmnZDl
📘 NLP Specialization bit.ly/3GXoQuP
✨ Free Resources that are great:
NLP: web.stanford.edu/class/cs224n/
CV: cs231n.stanford.edu/
Deployment: fullstackdeeplearning.com/
FastAI: www.fast.ai/
💻 My Deep Learning Setup and Recording Setup:
www.amazon.com/shop/aladdinpe...
GitHub Repository:
github.com/aladdinpersson/Mac...
✅ One-Time Donations:
Paypal: bit.ly/3buoRYH
▶️ You Can Connect with me on:
Twitter - / aladdinpersson
LinkedIn - / aladdin-persson-a95384153
Github - github.com/aladdinpersson
Timestamps:
0:00 Introduction
1:24 Why graphs
4:13 What is a graph
7:06 Common graph tasks
11:08 Representation of a graph
12:46 - How does a GNN work?
14:35 - Understanding information propagation
17:24 - Key property: Permutation Invariance
19:33 - Key property: Permutation Equivariance
22:22 - Message passing computation
23:53 - GNN Variant: Convolution
26:37 - GNN Variant: Attention
28:39 - Ending

[@AladdinPersson]

I think this video will be a great introduction for those of you who have heards about GNNs but never taken the time to learn what they actually are! It will give you the necessary knowledge so that you can follow more practical, hands-on coding using PyTorch Geometric in subsequent videos:)

[@sumithhh9379]

Are you going to publish a series on GNN and advanced concepts as well?

[@AladdinPersson]

@@sumithhh9379 It could become a series with more advanced topics but not sure. I've been working with GNNs through work projects and I felt I had enough knowledge to give a good introduction to the topic and show you how to use PyTorch Geometric to do the most common tasks (graph classification, link pred, node classification).

[@PhantomKenTen]

@@AladdinPersson I look forward to the PyTorch Geometric, I recommend using a molecular dataset

[@rembautimes8808]

A very good video especially the explanation on the computation graph. All the best on your startup’s success.

[@user-ij6mq7ze4z]

@AladdinPersson Thanks for this ; Helped the basics. Any part-2 of this?

[@prajwol_poudel]

Was actually waiting for you to make a series on GNNs . Can't wait for you to implement autoencoders/VAEs with more than just the innerproduct decoder.

[@guspretto]

I've been thinking and researching about GNNs for about a month. This video fit like a glove, great approach to the topic!

[@sharmakartikeya]

Hi Aladdin! Just came to drop a like and comment for the YT algorithm. Just love your videos.

[@AladdinPersson]

Much appreciated

[@starlite5097]

Thanks for the video, I never knew until now what GNNs do exactly. Please do such tutorials and paper implementations more often. I am currently watching your GAN playlist and it helps me a lot. Anything regarding GANs and semantic segmentation is most welcome!

[@quaterni0]

Hey Aladdin, just wanted to say thank you for this amazing content. Your explanations are clear and helpful and the way that you organize your videos makes me feel like I'm exploring these fascinating concepts right alongside you. All the best, brother.

[@anemone3409]

i really like the explanation of permutation invariance & equivariance, which i didn't understand when I read the textbook. super clear with those plots!

[@theaiframework1146]

Way to go. It is so great you can share the learnings from our projects

[@cwbh10]

Best explanation on GNNs on youtube right now dude, thank you so much

[@jibzys8946]

Great primer for getting a decent understanding of graphs. Thanks

[@736939]

Please continue this video series. GNN is the future with lot of applications. Please show how to create them from the scratch and how each type of GNN works (graphsage, GCN). Also the temporal GNN for prediction the messages that would be really cool!

[@tejasvix]

Bro waited for u for so long and got GNNs, highly excited for the series

[@pingalpratyushnath7786]

Great Job! You explained it quite well for someone who just heard the term. We wish more videos like this in the future.

[@user-vc6pi6kh8z]

We need a hands-on video😭 love your coding style so much!

[@edyu21]

This is very well done. Thank you. One thing that would be great would be explaining what happens when a and b are connected bidirectionally (nondirected) so that when computing a you'd use b and vice versa given t-1, t, t+1...

[@yandajiang1744]

This video is pretty clear and straightforward... a veeeerrry good start for beginners

[@filip4370]

Perfect timing mate 👌.

[@edpraveen]

Hi Aladdin! YT Notification popped up! Just came to put a comment and like. Love your videos!

[@mdbayazid6837]

Finally, legend is back. Love to see you again.

[@bennisyoutube]

I see the upload. I click on it, leave a like and then watch. Love your videos!

[@franciscoparraguez4576]

Amazing video, thank you so much. !!!!!!

[@morancium]

I always wanted this!!!

[@ten_tego_teges]

Great video, thanks!

[@ayazbaykov9646]

thanks, much needed

[@hwy9nightkid]

GNNs are the summer vibe

[@r3tex]

Bro, this is solid.

[@amiralizadeh6621]

looking forward to a video on unequal attention scores! thank you.

[@maximklechshev6675]

Please continue doing this

[@yb801]

Thanks for sharing this.

[@abiRahimSheikh]

Thank you!

[@ojussinghal2501]

Great explanation.

[@FLLCI]

Can't wait for the coding implementation!

[@yannickpezeu3419]

Thanks

[@RoyAAD]

Did you do the hands on?

[@BigAsciiHappyStar]

11:21 woohoo! I love puzzles ... no wait that's not a cryptic crossword 😭😭😭😭

[@SimonEliasen123]

I may be a bit shaky on the topic, but i think GNNs are only representation learning/embedding tools and do not perform tasks such as link prediction, node classification directly... These are only performed afterwards using separate neural network architecture. Please let me know if i am wrong, i am currently reading up on the topic and this specific point has led to much confusion on my part.

[@barathkandapan4794]

Hi @Aladdin, can you make a video on BERT and RoBERTa

[@MusixPro4u]

26:12 the k is the time step, isn't it? Not the layer.

[@SAINIVEDH]

17:05 Color must be purple instead of blue

[@Jake-om9no]

So where is the video about coding stuff using python's geometric? (doge

[@wireghost897]

Thanks a lot for your epxlaination but maybe next time prepare a script because the way you explain makes it look like you are making up stuff on the fly. I know that is not the case but it might give the impression that you don't know what you are talking about. Just an honest feedback.

[@AladdinPersson]

But I am making stuff up on the fly

[@AladdinPersson]

If you want I can use a larger beam size but might cause some video delay