Temporal Graph Networks (TGN) | GNN Paper Explained

❤️ Become The AI Epiphany Patreon ❤️ ► / theaiepiphany
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
A deep dive into the temporal graph networks paper.
You'll learn about:
✔️ What are dynamic graphs?
✔️ How to get a vectorized representation of time
✔️ All the nitty-gritty details behind the paper
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
✅ arxiv.org/abs/2006.10637
✅ Chris Olah on LSTMs: colah.github.io/posts/2015-08...
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
⌚️ Timetable:
00:00 Dynamic graphs
03:00 Suboptimal strategies
05:30 Terminology, temporal neighborhood
07:30 High-level overview of the system
08:35 We need to go deeper
13:30 Using temporal information to sample
14:10 Information leakage and the solution
16:55 Main modules explained
21:20 Memory staleness problem
24:00 Temporal graph attention
26:00 Vector representation of time
29:15 Batch size tradeoff
31:00 Results and ablation studies
33:55 Recap of the system
36:55 Some confusing parts
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
💰 BECOME A PATREON OF THE AI EPIPHANY ❤️
If these videos, GitHub projects, and blogs help you,
consider helping me out by supporting me on Patreon!
The AI Epiphany ► / theaiepiphany
One-time donation:
www.paypal.com/paypalme/theai...
Much love! ❤️
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
💡 The AI Epiphany is a channel dedicated to simplifying the field of AI using creative visualizations and in general, a stronger focus on geometrical and visual intuition, rather than the algebraic and numerical "intuition".
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
👋 CONNECT WITH ME ON SOCIAL
LinkedIn ► / aleksagordic
Twitter ► / gordic_aleksa
Instagram ► / aiepiphany
Facebook ► / aiepiphany
👨‍👩‍👧‍👦 JOIN OUR DISCORD COMMUNITY:
Discord ► / discord
📢 SUBSCRIBE TO MY MONTHLY AI NEWSLETTER:
Substack ► aiepiphany.substack.com/
💻 FOLLOW ME ON GITHUB FOR COOL PROJECTS:
GitHub ► github.com/gordicaleksa
📚 FOLLOW ME ON MEDIUM:
Medium ► / gordicaleksa
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
#temporalgraphnetworks #dynamicgraphs #graphml

[@user-co6pu8zv3v]

Thank you!

[@amirhosseindaraie5622]

Can you implement TGN?

[@alessandrobitetto2361]

Really nice video! What's the interpretation of inductive and transductive in the benchmark part?

[@TheAIEpiphany]

Check out my GAT jupyter notebook I've explained the difference there: github.com/gordicaleksa/pytorch-GAT/blob/main/The%20Annotated%20GAT%20(Cora).ipynb

[@DiogoSanti]

Need some background on NLP, but definitely coming back here in the near future...

[@TheAIEpiphany]

Yep unfortunately it's a universal problem, knowledge dependencies, in maths/engineering or teaching in general.

[@DiogoSanti]

​@@TheAIEpiphany Sure was too focused on ComputerVision, Yolo's and tracking stuff but theres a whole world out there... Started seeing some lessons from Stanford Cs224n, and reading this book so far: www.amazon.com/Transformers-Natural-Language-Processing-architectures-ebook/dp/B08S977X8K
Will definitely see your videos too to consolidate the subject, already subscribed here and followed you on Medium as you bring a lot of the cool stuff... Thanks for blessing community with your knowledge!

[@billykotsos4642]

So many NN architectures to get to grips with ... !!!

[@TheAIEpiphany]

That's true! Haha. It can get overwhelming. I am not there as well, but step by step. 😅

[@TheAIEpiphany]

TGN makes use of node memory information, node features, edge features, temporal information and last (but definitely not least) topological information.
Will you folks stop, please? Hahah