Stephane Chretien retweeted
✳️ Formation à l’#IA FIDLE du @CNRS. Elle est gratuite, en français, récente et très bien faite.
Via @capt_ad_hoc
youtube.com/@cnrs-fidle?si=A…
Stephane Chretien retweeted
Demain, à 16 heures sur @franceculture, nous ne ménagerons pas nos méninges, en compagnie de Richard Lévy.
Un cerveau, comment ça marche ? via @franceculture radiofrance.fr/franceculture…
Stephane Chretien retweeted
«A 2020 Vision of Linear Algebra» six brief videos for teaching and learning linear algebra by… yasss! Him! A 85 years old William Gilbert Strang! ❤️❤️❤️
I hope to have just a fraction of your stamina, dear Gilbert!
I’ll try to keep opening eyes!
ocw.mit.edu/resources/res-18…
Stephane Chretien retweeted
Je viens de numériser un livre d'Elias et Scotson assez dur à trouver si jamais ça vous intéresse. C'est une étude d'un quartier anglais qui essaye de comprendre sociologiquement comment se créent les logiques de groupe et d'exclusion.
Je vous mets un lien juste en dessous.
Stephane Chretien retweeted
There is no data-generating distribution. argmin.net/p/there-is-no-dat…
Stephane Chretien retweeted
All of my Deep RL course lecture videos from Spring 2025 are now online! 🥳
Youtube playlist: youtube.com/watch?v=EvHRQhMX…
Stephane Chretien retweeted
A new tokenizer is introduced for LLMs:
arxiv.org/pdf/2511.20849
Idea: Instead of merging tokens by frequency (BPE), optimize the tokenizer directly for maximizing average token length, yielding longer, more efficient tokens.
Results: 14–18% fewer tokens, faster training & inference, and better downstream accuracy across (small) GPT-style models.
Stephane Chretien retweeted
This is happening now in Ballroom AB #NeurIPS2025 !!
New paper on the generalization of Flow Matching arxiv.org/abs/2506.03719
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn **can only generate training points**?
with @Qu3ntinB, Anne Gagneux & Rémi Emonet 👇👇👇
Stephane Chretien retweeted
“So much of physics comes down to understanding geometry. And often in surprising ways.” —Jonathan Sorce, a theoretical physicist at Princeton University quantamagazine.org/what-is-a…
Stephane Chretien retweeted
Deep learning is not a collection of black-box tricks, contrary to what many believe. It can be learned as a principled engineering discipline.
This latest edition of Deep Learning with Python is my best attempt so far at teaching it. It focuses on building deep intuition -- the theory and mental models -- alongside all of the practical programming patterns.
You will learn this on a truly modern stack: Keras 3 as a high-productivity, framework-agnostic API, and preferably JAX for SotA performance and scalability (PyTorch and TensorFlow are also fully supported).
Stephane Chretien retweeted
Une lecture en soutien à Henry Laurens pour analyser les enjeux de l'annulation du colloque sur la Palestine & l'Europe au College de France : "Le Passé imposé" montre la nécessité de l'histoire pour dépasser l'affrontement des mémoires.
Vertigineux et plus que jamais nécessaire.
Stephane Chretien retweeted
New lectures posted on CS 2881 YouTube playlist youtube.com/playlist?list=PL…
Economic impacts with @RonnieChatterji & Bharat Chandar
Interpretabilty with @NeelNanda5, @Jack_W_Lindsey, @bobabowen & @nabla_theta
Stephane Chretien retweeted
Why and how does gradient/matrix orthogonalization work in Muon for training #LLMs?
We introduce an isotropic curvature model to explain it. Take-aways:
1. Orthogonalization is a good idea, "on the right track".
2. But it might not be optimal. [1/n]
Stephane Chretien retweeted
Everyone knows actions are fundamentally different than predictions, but it's hard to write this distinction in math. argmin.net/p/staging-interve…
Stephane Chretien retweeted
No matter what the models say, one person's statistical fact is another's statistical outlier. (Part 7 in the going-for-2-down-8 microcosm) argmin.net/p/learning-from-l…
Stephane Chretien retweeted
Amazing pictures (sound 100% on !) of the massive demonstration in Budapest and Hungarians' resolute "no" to populism and Russia yesterday 👇
... which didn't get even a mention in 𝒐𝒖𝒓 media, the same media which day in, day out are so keen to report on the march of populism
Stephane Chretien retweeted
new journal from the Association for Mathematical Research just dropped...
Stephane Chretien retweeted
The initial noise in diffusion models is surprisingly correlated with the final image.
Our NoisePrints paper exploits this to provide a lightweight, distortion-free, cryptographically secure watermark for proving authorship of generated images & videos, requiring no model access.
Stephane Chretien retweeted
Hi new followers! I’m a mathematician at Harvard. I have a YouTube channel, where I discuss math the way I think about it, with now two playlists on differential geometry and complex geometry (in progress). Comments welcome!
DG: youtube.com/watch?v=rVTN7VfE…
CG: youtube.com/watch?v=N5FQHgU6…



























