Researcher at @inria_Paris in the @Sierra_ML_Lab team
Joined September 2013
- Tweets117
- Following178
- Followers472
- Likes555
Pinned Tweet
Too busy/tired/lazy to find a convergence proof for your latest optimization algorithm? Let your computer do it!
PEPit is a new Python package for computer-assisted worst-case analyses (github.com/bgoujaud/PEPit or pypi.org/project/PEPit/).
(1/5)
Adrien Taylor retweeted
I surely repeat myself, but PEP is comfortably in the top 10 most exciting ideas which have been introduced (~) since I've been in research myself. The core questions are somehow very simple, but asked as ambitiously as possible: how to bound rates of convergence _sharply_?
This has been a thing in convex optimization since well before LLMs even existed! You can use optimization (solve a semidefinite program) to automatically generate a proof that another algorithm converges. These are called performance estimation problems, cc @TaylorAdrien
Adrien Taylor retweeted
I didn’t check the content but for this reason I am not sure we should trust Lean proofs more than human readable ones
arxiv.org/abs/2610.08144
Adrien Taylor retweeted
Too busy/tired/lazy to find a convergence proof for your latest optimization algorithm? Let your computer do it!
PEPit is a new Python package for computer-assisted worst-case analyses (github.com/bgoujaud/PEPit or pypi.org/project/PEPit/).
(1/5)
Adrien Taylor retweeted
This has been a thing in convex optimization since well before LLMs even existed! You can use optimization (solve a semidefinite program) to automatically generate a proof that another algorithm converges. These are called performance estimation problems, cc @TaylorAdrien
Adrien Taylor retweeted
arxiv.org/abs/2608.17252
Clever idea in this paper on "Mirror Polyak" that is analogous to the Polyak step size. Using ordinary Polyak is equiv. to a euclidean projection onto a half-space, instead Mirror Polyak does a Bregman projection.
Adrien Taylor retweeted
Poster today at 5pm! #ICML2026
"It's easier to tune the LR for method A than for B."
We tried to formalize this for model-based stochastic optimization methods.
We find a key quantity, called stability index, that describes how stable a (weakly) convex bound is as a function of LR.
📚arxiv.org/abs/2602.09842
Adrien Taylor retweeted
Very excited to be in Copenhagen for #EurIPS 🇪🇺
I am presenting an optimizer benchmark for diffusion model training (sunday @ PriGM workshop). it compares new methods (Muon, SOAP, ScheduleFree) to good old AdamW.
Happy to chat anytime ❄️
Adrien Taylor retweeted
Learning rate schedulers used to be a big mistery. Now you can just take a guarantee for *convex non-smooth* problems (from arxiv.org/abs/2310.07831), and they give you *precisely* what you see in training large models.
See this empirical study:
arxiv.org/abs/2501.18965
1/3
Adrien Taylor retweeted
The sudden loss drop when annealing the learning rate at the end of a WSD (warmup-stable-decay) schedule can be explained without relying on non-convexity or even smoothness, a new paper shows that it can be precisely predicted by theory in the convex, non-smooth setting!
1/2
Adrien Taylor retweeted
My book is (at last) out, just in time for Christmas!
A blog post to celebrate and present it: francisbach.com/my-book-is-o…
Adrien Taylor retweeted
Recruiting Post-Doctoral students in Machine-Learning, Optimization or Regulation in AI at TSE or Paris. Write to [email protected]
Adrien Taylor retweeted
We are looking for a post-doc in machine learning for astrophysics/astronomy ! pepr-origins.fr/2024/10/09/p…
Adrien Taylor retweeted
📰 #Article | Découvrez le projet ARTIFACT, porté par Justin Carpentier, responsable de l’équipe-projet commune WILLOW (@inria_paris @CNRSinformatics @ENS_ULM), qui vient d’obtenir une bourse #ERCStG. Félicitations Justin 🙌 !
👉 inria.fr/fr/bientot-des-robo…
[🧵...]
ALT Portrait de Justin Carpentier.
Adrien Taylor retweeted
📰 #Article | Découvrez le projet CASPER, porté par @TaylorAdrien, chercheur au sein de l'équipe-projet commune @Sierra_ML_Lab (@inria_paris /@ENS_ULM/@CNRSinformatics), qui vient d’obtenir une bourse #ERCStG. Félicitations Adrien 🙌 !
👉 inria.fr/fr/vers-des-algorit…
[🧵...]
ALT Portrait d'Adrien Taylor.
Adrien Taylor retweeted
🔴Félicitations à @julienmairal, lauréat du Prix @Inria-@AcadSciences Jeunes chercheurs🏆👏
Chercheur @inria_grenoble-@ugrenoblealpes, il explore les limites de l’apprentissage machine et de la vision par ordinateur dans de multiples domaines : inria.fr/fr/julien-mairal-ap…
A bit of random thoughts: the average number of daily arxiv submissions in computer science went from about 1.9 to 224.6 in 20 years.
It is probably not quite surprising that the reviewing system is being pressured?
Forgot to mention the source: arxiv.org/year/cs/22 (CS), arxiv.org/year/math/22 (math)
Adrien Taylor retweeted
Peut-on aimer les animaux et les manger?
C’est la question que s’est posée @GMeurice dans son nouveau livre aux éditions @ed_LaMartiniere.
À mettre entre toutes les mains! 👐
francebleu.fr/emissions/plan…
