Date/heure
24 septembre 2026
10:45 - 11:45
Lieu
Salle de conférences Nancy
Oratrice ou orateur
Victor Priser (Ponts et Chaussées)
Catégorie d'évènement Séminaire Probabilités et Statistique
Résumé
This talk is concerned with the long-time behavior of interacting particle systems of McKean–Vlasov type. We will first consider a general class of discrete-time stochastic particle systems and investigate their asymptotic behavior when both the number of particles and the time horizon tend to infinity. In particular, we will discuss how to obtain long-time convergence results beyond the classical uniform-in-time propagation of chaos framework, including situations where several McKean–Vlasov stationary measures may coexist. We will then turn to a specific application arising in zeroth-order optimization: Consensus-Based Optimization (CBO). CBO is a derivative-free optimization method in which particles interact through a weighted consensus point and are driven by noise. Our goal is to understand its behavior in the long-time regime and to establish theoretical guarantees for the finite-particle algorithm. In particular, we will quantify the different sources of error and study the interplay between the number of particles, the number of iterations, the discretization step, and the smoothing parameter.