♟️♟️Now our work on teaching superhuman chess strategies to grandmasters (one of whom Gukesh D who became the latest and the youngest world chess champion) is published on PNAS! 🎉🎉 Yes, we can transfer machine knowledge to humans to push the frontier of human. knowledge.

How does a diffusion model learn to mimic art styles? 🎨 Our latest work reveals that diffusion models create entirely new art styles to learn the concept - "art" 🤯 Checkout these art styles that Stability AI's SDXL has learnt. Do you recognize them?🤔 And we found more👇 🧵

Accepted at #ICML2025! Check out the preprint. Shoutout to the group for an AMAZING research journey Harry Thasarathan Julian Thomas Fel Matthew Kowal This is Harry’s first PhD paper (first year, great start) and Julian’s first ever paper (work done as an undergrad 💪).

Our work finding universal concepts in vision models is accepted at #ICML2025!!! My first major conference paper with my wonderful collaborators and friends Matthew Kowal Julian Thomas Fel Kosta Derpanis Working with y'all is the best 🥹 Preprint ⬇️

i am once again asking more people to do vision model interpretability

🚨 New paper alert! Linear representation hypothesis (LRH) argues concepts are encoded as **sparse sum of orthogonal directions**, motivating interpretability tools like SAEs. But what if some concepts don’t fit that mold? Would SAEs capture them? 🤔 1/11

Around CVPR for the next 2 days—if you're into interpretability, SAEs, complexity, or just wanna know how cool Kempner Institute at Harvard University is, hit me up 👋

*Universal Sparse Autoencoders* by Harry Thasarathan Thomas Fel Matthew Kowal Kosta Derpanis They train a shared SAE latent space on several vision encoders at once, showing, e.g., how the same concept activates in different models. arxiv.org/abs/2502.03714

Check out this amazing work by Fenil Doshi on shape holistic processing in vision models! 🍀

Chatted with Le Monde about interpretability and sparse autoencoders. (Yes, SAE made it into mainstream news 😅) lemonde.fr/pixels/article… Merci à Nicolas Six pour l’échange !

📢NEW POSITION PAPER: Use Sparse Autoencoders to Discover Unknown Concepts, Not to Act on Known Concepts Despite recent results, SAEs aren't dead! They can still be useful to mech interp, and also much more broadly: across FAccT, computational social science, and ML4H. 🧵

This was a really fun project to work on - and huge shoutouts to my amazing collaborators who made the project such a delight!! 🎉💪

What if adversarial examples aren't a bug, but a direct consequence of how neural networks process information? We've found evidence that superposition – the way networks represent many more features than they have neurons – might cause adversarial examples.

🕳️🐇Into the Rabbit Hull – Part I (Part II tomorrow) An interpretability deep dive into DINOv2, one of vision’s most important foundation models. And today is Part I, buckle up, we're exploring some of its most charming features.

Huge thanks to all collaborators who made this work possible, and especially to Binxu Wang 🐱. This work grew from a year of collaboration! Tomorrow, Part II: geometry of concepts and Minkowski Representation Hypothesis. 🕹️ kempnerinstitute.github.io/dinovision 📄 arxiv.org/pdf/2510.08638

Check out Multispectral Demosaicing via Dual Cameras #ICCV2025  Spotlight💡💡! In the future, cameras won’t just see color — they’ll read health, understand materials, and recognize life. Multispectral sensors are coming to your phone! Our work helps pave the way.

The most valuable output of research are the people not the papers. Thus the most important thing you can do is take care of your people.

Generative Point Tracking with Flow Matching My latest project with Adam W. Harley Kosta Derpanis (sabbatical @ CMU) Derek Nowrouzezahrai Chris Pal Project page: mtesfaldet.net/genpt_projpage/ Paper: arxiv.org/abs/2510.20951 Code: github.com/tesfaldet/genpt