AMLab

🤹 New blog post!

I write about our recent work on using hierarchical trees to enable sparse attention over irregular data (point clouds, meshes) - Erwin Transformer, accepted to ICML 2025

blog: maxxxzdn.github.io/blog/erwin/
paper: arxiv.org/abs/2502.17019

Compressed version in the thread below:

Don's miss this exciting opportunity! 🔥

🚨Application deadline on the 15th of June

One more week to apply to this exciting position... and another position on #CausalRepresentationLearning and #ReinforcementLearning for learning provably correct #concepts from raw data opening up soon!

Vacancy — PhD Position on Learning Concepts with Theoretical Guarantees Using Causality and RL Are you interested in improving the interpretability, robustness and safety of current AI systems? If the answer is yes, please continue reading!

New PhD position at the University of Amsterdam in @amlab.bsky.social on learning concepts with theoretical guarantees using #causality and #RL with me, Frans Oliehoek (TU Delft) and Herke van Hoof 💥

Deadline: 15 June

werkenbij.uva.nl/en/vacancies...

🐳 Controlled Generation with Equivariant Variational Flow Matching

By @eijkelboomfloor.bsky.social , @zmheiko.bsky.social ,
@sharvaree.bsky.social , @erikjbekkers.bsky.social ,
@wellingmax.bsky.social , @canaesseth.bsky.social *, @jwvdm.bsky.social *

📜 As above - paper TBA soon 😉

🧵8 / 8

🌊 Exponential Family Variational Flow Matching for Tabular Data Generation

By Andrés Guzmán-Cordero*, @eijkelboomfloor.bsky.social *,
@jwvdm.bsky.social

📜 This paper will be shared soon - keep your eyes open! 🤩

🧵7 / 8

⚠️ The Perils of Optimizing Learned Reward Functions: Low Training Error Does Not Guarantee Low Regret

By Lukas Fluri*, @leon-lang.bsky.social *, Alessandro Abate, Patrick Forré, David Krueger, Joar Skalse

📜 arxiv.org/abs/2406.15753

🧵6 / 8

🔁 SDE Matching: Scalable and Simulation-Free Training of Latent Stochastic Differential Equations

By @gbarto.bsky.social , Dmitry Vetrov, @canaesseth.bsky.social

📜 arxiv.org/abs/2502.02472

🧵5 / 8

📐 On the Importance of Embedding Norms in Self-Supervised Learning

By Andrew Draganov, @sharvaree.bsky.social , Sebastian Damrich, Jan Niklas Böhm, Lucas Maes, Dmitry Kobak, @erikjbekkers.bsky.social

📜 arxiv.org/abs/2502.09252

🧵4 / 8

⚖️ A machine learning approach to duality in statistical physics

By Prateek Gupta, Andrea Ferrari, @nabiliqbal.bsky.social

📜 arxiv.org/abs/2411.04838

🧵3 / 8

🐸 Erwin: A Tree-based Hierarchical Transformer for Large-scale Physical Systems

By @maxxxzdn.bsky.social , @jwvdm.bsky.social , @wellingmax.bsky.social

📜 arxiv.org/abs/2502.17019

🧵2 / 8

Exciting news: AMLab is happy to have 7 papers accepted at #ICML2025! 🎉

See the thread below for the full list 📝 and meet us in Vancouver to discuss them further! 🇨🇦

🧵1 / 8

🥇 3 Best Paper Awards for AMLab members!

1. Towards Variational Flow Matching on General Geometries by @olgatticus.bsky.social et al.

2. Generative Uncertainty in Diffusion Models by @metodjazbec.bsky.social et al.

3. SDE Matching by @gbarto.bsky.social et al.

Congrats to everyone! 🔥

Test of Time Winner

Adam: A Method for Stochastic Optimization
Diederik P. Kingma, Jimmy Ba

Adam revolutionized neural network training, enabling significantly faster convergence and more stable training across a wide variety of architectures and tasks.

🤹 Excited to share Erwin: A Tree-based Hierarchical Transformer for Large-scale Physical Systems

joint work with @wellingmax.bsky.social and @jwvdm.bsky.social

preprint: arxiv.org/abs/2502.17019
code: github.com/maxxxzdn/erwin

A few weeks ago, I presented SNAP at the wonderful #Bellairs Workshop on Causality in Barbados🐢

This Friday 🫰meets 🤌 as I will get to present SNAP again at the kick-off of the newest season of @causalclub.di.unipi.it! Check out this, and their other amazing upcoming talks at causalclub.di.unipi.it

Variational Flow Matching goes Riemannian! 🔮

In this preliminary work, we derive a variational objective for probability flows 🌀 on manifolds with closed-form geodesics, and discuss some interesting results.

Dream team: Floor, Alison & Erik (their @ below) 💥

📜 arxiv.org/abs/2502.12981
🧵1/5

Congratulations to "my" first PhD student Dr. @rmassidda.it who defended today at the University of Pisa *with honours* a thesis on "Methodological Advancements for Causal Abstraction Learning" 🎉

Riccardo is an amazing @ellis.eu PhD co-supervised w/ Davide Bacciu
@ellisamsterdam.bsky.social

💡 ➡️ 🧠 ➡️ 🚀

😍💥

Facts 💃🏻

The Christmas spirit has arrived at AMLab! 🎄✨
Yesterday we kicked off the holidays with a festive group dinner and a fun Secret Santa exchange. 🎅

Wishing everyone a restful and joyful winter break and a happy new year! ❄️💫

looks can be deceiving! ;)

Professor Imbens also had a mentoring session with our PhD students actively working on causality, discussing their ideas and the potential impact of their applications! 👨‍🔬👩‍🔬

@matyasch.bsky.social @roelhulsman.bsky.social @rmassidda.it @danruxu.bsky.social 🔥

Yesterday, we had the honor of hosting 2021 Nobel Laureate in Economics, Guido Imbens, at our Lab! 🤩

We had the chance to attend his inspiring talk on Experimental Design in Marketplaces 📈👩‍💻

A big thank you to @smaglia.bsky.social for making this event possible 💥

🧑‍🏫Neural Flow Diffusion Models at #NeurIPS2024 tomorrow! Discover how to build learnable noising processes for straight-line generative trajectories end-to-end and without simulations!🤯

📍West Ballroom A-D #6809
⏰Fri 13 Dec 4:30 pm — 7:30 pm

🔗https://neurips.cc/virtual/2024/poster/94656

Here are the dapper @metodjazbec.bsky.social and Alexander Timans, presenting their work on early-exit with risk control.

Come see @eijkelboomfloor.bsky.social and @gbarto.bsky.social present their work on variational flow matching now in West Ballroom A-D #7103!

François Cornet and @gbarto are presenting their work on Equivariant Neural Diffusion now at East Exhibition #2403!

Come talk to @zmheiko.bsky.social about his work on sample-efficient black box variational inference! (East Exhibition Hall #4105)