
Activités scientifiques du département
Le DMA est à la fois un département d'enseignement et un département de recherche. Cette structuration originale vise notamment à mettre très tôt les élèves au plus près de la recherche en train de se faire.
Publications
L'essentielle de publications des membres du département, des thèses et des HDR qui y sont soutenues sont disponibles sur le serveur HAL.
-
20 November 2026 hal-05874451 publication
For deterministic monotone cellular automata on the d-dimensional integer lattice, Toom has given necessary and sufficient conditions for the all-one fixed point to be stable against small random perturbations. The proof of sufficiency is based on an intricate Peierls argument. We present a simplified version of this Peierls argument. Our main motivation is the open problem of determining stability of monotone cellular automata with intrinsic randomness, in which for the unperturbed evolution the local update rules at different space-time points are chosen in an i.i.d. fashion according to some fixed law. We apply Toom's Peierls argument to prove stability of a class of cellular automata with intrinsic randomness and also derive lower bounds on the critical parameter for some deterministic cellular automata.
Jan M Swart, Réka Szabó, Cristina Toninelli
-
20 November 2026 hal-05873741 pré-publication
Let G be a group with a non-elementary action on a (not necessarily discrete) A~2-buildings. We prove that, given a random walk on G, isometries in G are strongly regular hyperbolic with high probability. As a consequence, we prove a Tits alternative for G, as well as a local-to-global fixed point result. We also prove that isometries of (not necessarily complete) R-buildings are semi-simple.
Corentin Le Bars, Jean Lécureux, Jeroen Schillewaert
-
15 October 2026 tel-05801443 thèse
This PhD thesis presents contributions to the field of deep learning. From convolutional ResNets to Transformers, residual connections are ubiquitous in state-of-the-art deep learning models. The continuous depth analogues of residual networks, neural ODEs, have been widely adopted, but the connection between the discrete and continuous models still lacks a solid mathematical foundation. In this manuscript, we will show that for a formal correspondence between residual networks and neural ODEs to hold, the residual functions must be smooth with depth, and we will present an implicit regularization result of deep residual networks towards neural ODEs. We will then present two applications of this analogy to the design and study of new architectures. First, we will introduce a drop-in replacement for any residual network that can be trained with the same accuracy, but with much less memory. Second, by viewing the attention mechanism as an interacting particle system, where the particles are the tokens, we will study the impact of attention map normalization on the Transformer model. Finally, we will present some other contributions to Transformers: how Transformers perform in-context autoregressive learning and how to differentiably route tokens to experts in Sparse Mixture of Experts Transformers.
Michaël Eli Sander
Les actualités de la recherche
Annonce de conférences, congrès et autres événements scientifiques.
-
Lancement du séminaire math-environnement au DMA
A partir du 14 octobre, ce nouveau rendez-vous, créé par le comité environnement du département,
-
Serre 100, une conférence en l’honneur du centième anniversaire de Jean-Pierre Serre à l’IHP
Rediffusion sur la chaîne youtube de l'IHP de la conférence du mardi 15 et mercredi
-
OpenAI annonce une solution à l’équation de Navier-Stokes : deux membres du DMA interviennent sur Radio France
Gabriel Peyré et Amaury Hayat ont été interrogé pour contextualiser cette événement majeur en

Annales de l’ENS
Les Annales scientifiques de l’École normale supérieure publient 6 fascicules par an. Elles sont éditées par la Société mathématique de France depuis 2008.

