| Les deux révisions précédentes
Révision précédente
Prochaine révision
|
Révision précédente
|
start [2026/01/13 10:14]
Igor Kortchemski |
start [2026/05/06 11:44] (Version actuelle)
Djalil Chafaï |
| problèmes connexes. | problèmes connexes. |
| // | // |
| * 16 mars 2026. **[[https://marylou-gabrie.github.io/|Marylou Gabrié (LPENS)]]**. **Titre à préciser.**\\ //// | * 16 mars 2026. **[[https://marylou-gabrie.github.io/|Marylou Gabrié (LPENS)]]**. **De l’utilisation des modèles génératifs en échantillonnage.**\\ //Les modèles génératifs profonds paramètrent des familles de distributions très flexibles, capables de representer des ensembles de données complexes, tels que des images ou du texte. Ces modèles fournissent des échantillons indépendants provenant de distributions complexes de haute dimension à un coût négligeable. En revanche, échantillonner exactement une distribution cible, comme une loi posterior Bayésienne ou la distribution de Boltzmann d’un système physique, est généralement difficile : soit en raison de la dimensionnalité, de la multimodalité, du mauvais conditionnement, soit d’une combinaison de ces facteurs. Dans cet exposé, je discuterai des opportunités et des défis liés à l’amélioration des algorithmes traditionnels d’inférence et d’échantillonnage à l’aide de l’apprentissage.// |
| * 13 avril 2026. **[[https://www.sertedonderwinkel.com/|Serte Donderwinkel (Groningen, Professeure invitée ENS-PSL/DMA)]]**. **Titre à préciser.**\\ //// | * 13 avril 2026. **[[https://www.sertedonderwinkel.com/|Serte Donderwinkel (Groningen, Professeure invitée ENS-PSL/DMA)]]**. **Counting connected graphs.**\\ //How many connected graphs have a prescribed degree sequence?
This classical combinatorial question turns out to admit a natural probabilistic approach. |
| * 11 mai 2026. **[[https://www.ceremade.dauphine.fr/~massoulie/|Brune Massoulié (Dauphine)]]**. **Titre à préciser.**\\ //// | |
| | In joint ongoing work with Sasha Bell and Remco van der Hofstad, we derive asymptotic formulas for the number of connected graphs with a given degree sequence. Our approach is an example of the probabilistic method: rather than counting directly, we introduce a suitable random graph model and study the likelihood that it exhibits a desired structure. |
| | |
| | Concretely, we construct a random graph in which (an approximation of) the prescribed degree sequence appears with high probability inside a large connected component. This perspective allows us to translate questions about enumeration into probabilistic statements about random graphs. |
| | |
| | Along the way, I will discuss several key probabilistic tools, including the configuration model, branching process approximations, and local weak convergence, and explain how they combine to yield asymptotic counting results.// |
| | * 27 avril 2026. ** [Attention, exceptionnellement à 14h] [[https://www.robinkhanfir.com/|Robin Khanfir (McGill)]]**. **The Brownian tree is the only uniformly self-similar binary tree.**\\ //The Brownian tree is the scaling limit of many random tree models for which the square of the diameter is of the order of the number of vertices. In contrast to this universality, proofs of such convergences commonly rely on model-specific methods. To provide a conceptual understanding of the universality of the Brownian tree, we show that it is uniquely characterized by a uniform self-similar decomposition property. This leads to a general proof scheme for convergences to the Brownian tree that does not require the computation of finite-dimensional limit distributions. This talk is based on a work in progress.// |
| | * 11 mai 2026. **[[https://www.ceremade.dauphine.fr/~massoulie/|Brune Massoulié (Dauphine)]]**. **An introduction to some self-repelling processes.**\\ //Self-repelling walks and processes are stochastic processes that are influenced by their past behaviour, in a way that makes them try to avoid their past trajectory. In this talk, I will first present a toy model for self-repelling random walks introduced by Toth and Werner, which allows to present results and methods that generalise to more complex models. I will then present the « true » self-avoiding walk (TSAW) and state the results from an article by Toth in 1995. Last, I will informally present the « true » self-repelling motion, which was constructed by Toth and Werner in 1998, and was proved to be the limit of the TSAW very recently by Kosygina and Peterson.// |
| * **Année 2024-2025.** | * **Année 2024-2025.** |
| * **Organisateurs[[organisation|.]]** [[https://djalil.chafai.net/|Djalil Chafaï]] et [[https://www.normalesup.org/~dumaz/|Laure Dumaz]] | * **Organisateurs[[organisation|.]]** [[https://djalil.chafai.net/|Djalil Chafaï]] et [[https://www.normalesup.org/~dumaz/|Laure Dumaz]] |