New insights on the learning of Gaussian mixtures in high dimension in a new preprint by Bruno Loureiro, Gabriele Sicuro, Cedric Gerbelot, A. Pacco, Krzakala Florent, and Lenka Zdeborova. Rigorous results, replica computation and test on real datasets arxiv.org/abs/2106.03791 #KCLDisSystResearch

1st: A proof of state evolution for an impressively broad class of approximate message passing algorithms, including multilayer ones, generative priors etc. in "Graph-based Approximate Message Passing Iterations" by Cedric Gerbelot and Raphael Berthier arxiv.org/pdf/2109.11905…

The paper on Gaussian mixtures learning by Bruno Loureiro, Gabriele Sicuro, Cedric Gerbelot, A. Pacco, Krzakala Florent and Lenka Zdeborova has been accepted to NeurIPS Conference as spotlight (top 3% paper)! Check out the final version here bit.ly/3Dqzarz #KCLDisSystResearch

📜Using ensembling to demystify fluctuations and correlations in GLMs trained on different, but correlated, features, in a new preprint by Bruno Loureiro, Cedric Gerbelot, Maria Refinetti, Gabriele Sicuro, and Krzakala Florent. On the ArXiv: arxiv.org/abs/2201.13383 #KCLDisSystResearch

Mathematicians will unironically say things like "For a concrete example, consider a smooth function f on the interval [a,b]"

The state evolution of approximate message passing can also be derived and proven for random convolutional matrices. This gives theoretical results for inverse problems with convolutional generative priors. Kudos to Max and Cedric Gerbelot arxiv.org/abs/2205.13503

They grow up so fast… good luck to my soon-former student Cedric Gerbelot for his PhD defense. And all the best for your postdoc at NYU Courant!

#OpenCall| Félicitations à Bruno Loureiro Bruno Loureiro recruté dans le cadre du programme #ChooseFrance du CNRS 🌍. Il intégrera le #DIENS pour poursuivre ses travaux en apprentissage automatique et physique statistique. ➡️ ins2i.cnrs.fr/fr/cnrsinfo/ch… 🤝CNRS Paris-Centre École normale supérieure | PSL Inria

It is satisfying when students outgrow me and pull off work that I did not think feasible only a year ago. Cedric Gerbelot 's proof of the dynamical mean field theory also in discrete time and for stochastic updates is a prime example of that: arxiv.org/pdf/2210.06591…

Physicists often think of Machine Learning Algos as minimizing the energy of a physical system & use big guns, such as dynamical mean-field theory, to study the learning dynamics analytically. With Cedric Gerbelot, Lenka Zdeborova, F.Mignacco & E.Troiani, we bring rigor to these ideas

Dear Friends, please share broadly the announcement of the workshop on "Statistical Physics & Machine Learning Back Together Again" in Cargese 31.7.-12.8. 2023. A list of confirmed speakers and application form can be found at: cargese2023.github.io (deadline 31st March)

Interested in coming for a postdoc at Ecole Normale Supérieure in Paris? To work in statistical and theoretical physics? With a very prestigious position? Check these offers and apply! academicjobsonline.org/ajo/jobs/28737 academicjobsonline.org/ajo/jobs/27869