Can one model generalize to unseen graphs with arbitrary feature and label spaces?.📢 Introducing GraphAny #ICLR25: A fully-inductive model for node classification. 🚀 Trained on one graph, GraphAny generalizes to any unseen graph. 📎 Paper: 1/🧵

RT @gsp_workshop: 🚨 The 8th annual Graph Signal Processing Workshop is back this May 14-16! Held in Montreal, CA, at @Mila_Quebec, we’re co….

🙌 Shoutout to the amazing team: @zhu_zhaocheng, @michael_galkin, Hesham Mostafa, @mmbronstein, and @tangjianpku! Working with such brilliant minds has been an amazing journey. See you in Singapore!. Check out the code and test on your own graph at: 6/6.

🚀 Efficiency:. You can train GraphAny on 1 graph and run inference on 30 graphs—with a total of 3 million nodes and 244 million edges—all within 10 minutes. Perform experiments on a single V100 or even a MacBook CPU!. 5/🧵

Inductive Attention:. GraphAny learns inductive attention scores for each node to fuse the predictions of multiple LinearGNNs. It adaptively predicts the important LinearGNNchannels via transforming the distances features between LinearGNNs. 4/🧵

Inference on Any Graph:. GraphAny models inference on a new graph as analytical solutions to LinearGNNs, thereby enjoys the training-free inference on arbitrary feature and label spaces. 3/🧵

GraphAny trained on only one graph (Wisconsin with 120 labeled node) effectively generalizes to 30 new graphs with an average accuracy of 67.26% in a fully-inductive manner, surpassing GCN and GAT individually trained in the transductive regime. 2/🧵

RT @xiaoyongzhu001: I'm so surprised and offended to see the professor mention "Chinese students are dishonest" during one of the invited t….

RT @Mila_Quebec: Mila's annual supervision request process opens on October 15 to receive MSc and PhD applications for Fall 2025 admission!….

RT @olexandr: Cats are invariant under SO(3) transformations! 😼

RT @michael_galkin: 📣Foundation models for graph reasoning become even stronger - in our new #NeurIPS2024 work we introduce UltraQuery goin….

RT @svpino: This thread is just ~30% of the full post, which you can find on Tivadar's book. You won't find better explanations anywhere e….

RT @michael_galkin: 📢 In our new blogpost w/ @mmbronstein @haitao_mao_ @AndyJiananZhao @zhu_zhaocheng we discuss foundation models in Graph….

RT @elonmusk:

RT @gregisenberg: How to stay calm (without any hacks) . This is all you need.

RT @rohanpaul_ai: This is really a 'WOW' paper. 🤯. Claims that MatMul operations can be completely eliminated from LLMs while maintaining s….

RT @HamelHusain: LLM bullshit knife, to cut through bs. RAG -> Provide relevant context.Agentic -> Function calls that….

lol.

RT @leopoldasch: Virtually nobody is pricing in what's coming in AI. I wrote an essay series on the AGI strategic picture: from the trendl….

RT @lexfridman: "Think lightly of yourself and deeply of the world." - Miyamoto Musashi.