For those asking for material - I will mostly be following the amazing "elements of differentiable programming" (arxiv.org/abs/2403.14606) by Mathieu Blondel Vincent Roulet with some added labs & material. 🙃

Twitter friends, help me share! Sapienza Università di Roma has an incoming PhD deadline approaching - contact me if you want more info and/or if the idea of a PhD with me intrigues you (not mutually exclusive). 🙃 [Disclaimer: The sadness of our stock photo is not representative of PhD life]

overseen in the recent GNN+ paper: gnn arch diagrams that are remarkably transformer-style. this might help make gnns more approachable, and is appreciated. the results are cool too! it further highlights nicely how gnns/transformers have compatible underlying symmetries :)

Happy to share I just started as associate professor in Sapienza Università di Roma! I have now reached my perfect thermodynamical equilibrium. 😄 Also, ChatGPT's idea of me is way infinitely cooler so I'll leave it here to trick people into giving me money.

Excited to unveil Boltz-2, our new model capable not only of predicting structures but also binding affinities! Boltz-2 is the first AI model to approach the performance of FEP simulations while being more than 1000x faster! All open-sourced under MIT license! A thread… 🤗🚀

Delighted to announce the release of Boltz-2, which demonstrates unprecedented accuracy in predicting structure and binding affinity! Congrats to Gabriele Corso and Jeremy Wohlwend on this breakthrough achievement! 📄Paper: bit.ly/boltz2-pdf 💻Code: github.com/jwohlwend/boltz

Great to see our work out in Angewandte Chemie! 🎉 We introduce ‘supramolecular’ language processing to predict co-crystallization — with experimental validation! 🙈 Led by the unstoppable Rebecca Birolo, w/ Rıza Özçelik; TU Eindhoven, Molecular Machine Learning; Università di Torino 💪🏻 onlinelibrary.wiley.com/doi/10.1002/an…

#NowPlaying open.spotify.com/album/51B7LbLW…

*Interpretability in Parameter Space* by Dan Braun Lee Sharkey & al. They look for "interpretable" components directly in weight space of a model, by optimizing for several desiderata (faithfulness, low-rank, etc.). For now only works on toy tasks. arxiv.org/abs/2501.14926

pouring my heart into this chapter's visuals 😊

the 1st draft 'g' chapter of the geometric deep learning book is live! 🚀 alice enters the magical, branchy world of graphs & gnns 🕸️ (llms are there too!) i've spent 7+ years studying, researching & talking about graphs. this text conveys what i've learnt. more in thread 💎

Twitter friends, here's some draft notes for my upcoming course on automatic differentiation, mostly based on the "Elements of differentiable programming" book. Let me know what you think! They also include a notebook on operator overloading. 🙃 notion.so/sscardapane/Au…

one for my theory friends: filter equivariant functions

*Into the land of automatic differentiation* Material is out! A short PhD course for the CS PhD in Sapienza Università di Roma covering basic and advanced topics in autodiff w/ slides, (rough) Notion notes, and two notebooks including a PyTorch-like implementation. 😅 sscardapane.it/teaching/phd-a…

*Dense Backpropagation Improves Training for Sparse MoEs* by Ashwinee Panda Tom Goldstein et al. They modify the top-k router of a MoE by adding a "default" activation for unselected experts in order to have a dense gradient during the backward pass. arxiv.org/abs/2504.12463

😎

In the 20th century, non-Euclidean geometry transformed how we model the world with pen&paper. Now, it disrupts how we model the world w machines👇 Mathilde Papillon🦋 mathildepapillon .bsky .social Sophia Sanborn Johan Mathe Louisa Cornelis Abby Bertics Buracas Hansen Lillemark Christian Shewmake Fatih Dinc Pennec

#leadership

😎