🚨 ICML 2025 Paper 🚨 "On Measuring Long-Range Interactions in Graph Neural Networks" We formalize the long-range problem in GNNs: 💡Derive a principled range measure 🔧 Tools to assess models & benchmarks 🔬Critically assess LRGB 🧵 Thread below 👇 #ICML2025

🚨 Exciting news! We released 🎱 tgp (Torch Geometric Pool), the library for pooling in Graph Neural Networks. 🚀 Get started with our tutorials: https://t.co/Zhkp551VVv With @IvanMarisca and Carlo Abate. #GraphML #GNN #Pooling #Pyg

Freshly accepted as a tutorial paper in ACM Computing Surveys! 🥳 Check out the updated version: https://t.co/bboAZQSFmA cc @IvanMarisca @dan_zambon

🔥 Bayesian nonparametrics meets pooling in GNNs! 🔥 This is the first clustering-based pooling method that learns and dynamically adapts the size of the pooled graph to input and downstream task. 📄 paper: https://t.co/SwGSHLyFzE 💻 code: https://t.co/WxjydvFaIz

3. Graph Deep Learning for Time Series Processing: Forecasting, Reconstruction, and Analysis by @andreacini1994, @IvanMarisca, @dan_zambon 4. Integrating Knowledge Graphs and Large Language Models for Advancing Scientific Research by @qzhang_cs, @ChenJiaoyan1, @mengzaiqiao

BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Chemistry with one half to David Baker “for computational protein design” and the other half jointly to Demis Hassabis and John M. Jumper “for protein structure prediction.”

A pleasure co-advising such great students 👏

In a few hours, I will present HiGP at #ICML2024 😊 If you are interested in how to exploit graph-based representations and hierarchical clustering to forecast correlated time series, come and say hi at poster #907 from 1:30 to 3:00 PM 😉

Come to visit me and @IvanMarisca tomorrow at the first poster session (11:30-13:00 - Hall C) as we present our paper #1016! #ICML2024

@FilippoMariaBi1 5/ Paper and code available here: https://t.co/8rzZX72uyq

@FilippoMariaBi1 4/ Two bonus features: ⚡️ The scalability is given by factorized time-then-space processing and precomputed downsampling operators. 🔍 Gain insights on the scales through the scores! 👇

@FilippoMariaBi1 3/ 🚀 We propose HD-TTS, an STGNN for forecasting with missing data through spatiotemporal downsampling. It follows 2 steps: Compute a hierarchy of representations at different spatiotemporal scales. Adaptively weigh representations depending on the input missing data pattern.

@FilippoMariaBi1 2/ To recover corrupted dynamics, we need diverse processing for the different space-time scales. However, multi-scale processing might affect scalability in both time and space.

@FilippoMariaBi1 1/ Why? Graph Deep Learning excels in modeling spatial dependencies for time series forecasting. Missing data – common in real-world settings – corrupts input dynamics, and most GNN predictors require complete sequences as input.

🧵 Ready for #ICML2024! This year me and @FilippoMariaBi1 present a method to forecast correlated time series with missing data. We compute a hierarchy of multi-scale spatiotemporal representations and adaptively combine them conditioned on the missing data pattern👇

A dinner with old, current, and future students and postdocs

Among other things, #NorthernLights were definitely something I didn't see coming during my stay in #Oxford (~1h ago)

Visit your new scholar profile page! https://t.co/XNKTw7LPrg BetterScholar aims to be a better alternative to Google Scholar to quickly get an overview of a scientist. It is built from scratch starting from 5 principles: 🧵

It's official: I'm now an ELLIS (@ELLISforEurope) PhD student! 🎉 The ELLIS PhD program aims to connect outstanding young ML researchers in Europe. Thanks to professors Cesare Alippi and Michael Bronstein (@mmbronstein) for being my advisors on this journey 🙏

Excited to embark on a 6-month journey at @UniofOxford with Prof. @mmbronstein and his group! Looking forward to working on graphs in such an amazing environment 👨‍🚀🚀