🔥Telegram İfşa

Afraid of sum-of-squares (SOS) relaxations? Read this new blog post for a smooth ride in the Fourier domain. francisbach.com/sums-of-square…

Afraid of Nesterov acceleration? Build an intuition from (almost) only illustrations, in this month blog post. francisbach.com/continuized-ac…

A quiz, code snippets, and strong duality without convexity. What could be more fun than that? See this month’s blog post on non-convex quadratic optimization on the sphere. francisbach.com/non-convex-qua…

Conjugate gradients are commonly used for solving linear systems, but their generalized nonlinear variants do not seem popular these days. Looks like for a good reason: in the worst case, they are not even better than gradient descent. arxiv.org/abs/2301.01530

Ben Grimmer Shameless plug: last year we investigated cyclical learning rate schedules with momentum. We found out that under some assumptions it can even go probably faster than classical accelerated methods. Blog post: fa.bianp.net/blog/2022/cycl… and paper: arxiv.org/abs/2106.09687

Heavy-ball method does not necessarily provide accelerated convergence on smooth and strongly convex objectives. Cool!

Excellent paper and surprising result from @AymericD4, B.Goujaud, and Adrien Taylor: while the heavy-ball method provably accelerates optimization on quadratic functions, it does not accelerate for all smooth and strongly convex functions!

📚 #ÉtéLecture | Piquez une tête dans nos archives ! Aujourd'hui : Comment améliorer les capacités de locomotion 🦿 et de manipulation 🦾 des #robots 🤖 ? Découvrez Pinocchio, un logiciel #opensource conçu et développé par Justin Carpentier (WILLOW). inria.fr/fr/pinocchio-l…

Peut-on aimer les animaux et les manger? C’est la question que s’est posée GMeurice dans son nouveau livre aux éditions Éditions de La Martinière. À mettre entre toutes les mains! 👐 francebleu.fr/emissions/plan…

Académie des sciences Centre Inria de Saclay @ENS_ParisSaclay Dassault Systèmes FR Centre Inria d'Université Côte d'Azur Université de Bordeaux 🔴Félicitations à Julien Mairal, lauréat du Prix Inria-Académie des sciences Jeunes chercheurs🏆👏 Chercheur Centre Inria de l'Université Grenoble Alpes-@ugrenoblealpes, il explore les limites de l’apprentissage machine et de la vision par ordinateur dans de multiples domaines : inria.fr/fr/julien-mair…

📰 #Article | Découvrez le projet ARTIFACT, porté par Justin Carpentier, responsable de l’équipe-projet commune WILLOW (Centre Inria de Paris CNRS Sciences informatiques École normale supérieure | PSL), qui vient d’obtenir une bourse #ERCStG. Félicitations Justin 🙌 ! 👉 inria.fr/fr/bientot-des… [🧵...]

📰 #Article | Découvrez le projet CASPER, porté par Adrien Taylor, chercheur au sein de l'équipe-projet commune Sierra Team (Centre Inria de Paris /École normale supérieure | PSL/CNRS Sciences informatiques), qui vient d’obtenir une bourse #ERCStG. Félicitations Adrien 🙌 ! 👉 inria.fr/fr/vers-des-al… [🧵...]

We are looking for a post-doc in machine learning for astrophysics/astronomy ! pepr-origins.fr/2024/10/09/pos…

Recruiting Post-Doctoral students in Machine-Learning, Optimization or Regulation in AI at TSE or Paris. Write to [email protected]

My book is (at last) out, just in time for Christmas! A blog post to celebrate and present it: francisbach.com/my-book-is-out/

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

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