We're talking with this guy tomorrow. Come join the discussion of you are into graph representation learning!

Come say hi. We are presenting our paper Directional Graph Neural Networks tomorrow at ICLR GTRL workshop.

Try doing message passing along the gradient of the eigenfunction of the graph Laplacian. It reduces over smoothing! It makes your network learn better by breaking the symmetry of the nodes neighborhood! More ont this at ICML 2021

Check this out. Hypergraphs and simplicial complexes are the future. Graphs are great but they are obviously very limited and by design cannot capture higher order relationships.

graph transformers are getting better! In arXiv:2012.09699 the Laplacian eigenvectors are already used as positional encoding but here we show passing this info to a learned positional encoder along with preserving connectivity info improves performance

Join the GraphML reading group this Tuesday for "Rethinking Graph Transformers with Spectral Attention" arxiv.org/abs/2106.03893 presented by authors Dominique Beaini Devin Kreuzer! With WL. Hamiliton Vincent Létourneau Prudencio M Tossou The Zoom is here: hannes-stark.com/logag-reading-… 1/2

Spectral graph transformers are GOOOOOOOOOOOOOOOOOOOOO!

Really cool and surprising work

Come talk with us at the poster session of Neurips workshop on ML and physics x.com/dom_beaini/sta…

[🧠 collective intelligence 🧠] I've been intrigued by the cellular automata (CA) concept for a long time, and by the potential, mutually beneficial interaction between CAs and deep learning, so I decided to dig a bit deeper. Here are some interesting resources I found: 1/🧵

Equifax is selling access to people's salary data as part of an "employment verification solution" that includes: Previous annual salaries Previous weekly paycheck amounts Previous addresses Who accessed the report in the last 24 hours Etc. news.ycombinator.com/item?id=298347…

Twitter is so frustrating to navigate! I just saw this amazing tweet, hit the wrong button, hit back, cant find the tweet again in my feed! how can this happen? how to prevent this??

Mindblowing

Really fun discussion

This Tuesday in LoGaG Francesco Di Giovanni Jake Topping and @b_p_chamberlain present their "Understanding over-squashing and bottlenecks on graphs via curvature" arxiv.org/abs/2111.14522 - an absolute super-star paper from #iclr2022!🚀 Join the meeting here: hannes-stark.com/logag-reading-… 1/2

I love this, I'm betting there is a lot more ways discrete differential geometry can be used to make sense of gnns

Electric cars are cool but they seem to have a lot of serious flaws like having to rebuild the distribution system and reliance on some rare minerals. Making fuel from algae seems important to explore

🧵1/3 In the past 2 years, models posted on the OGB-molHIV datasets saw their test score increase, but val decrease. Models never stay on top of benchmarks for long. But, I'm proud to say that our models, PNA and DGN, have held the top spot on the val roc-aucfor almost 2 years!

CoLLAs (lifelong learning) conference starts tomorrow!

📣 We’re excited to announce the Graph Positional and Structural Encoder (GPSE) 🌐: The first graph encoder that learns highly transferable, unified representations for multiple positional encodings! 📜 Paper: arxiv.org/abs/2307.07107 🛠 Code: github.com/G-Taxonomy-Wor… 👇 🧵[1/11]