Sampling small noise dynamical rare events: the fluctuations of splitting Monte Carlo methods

par Mathias Rousset (INRIA Rennes)

mardi 2 déc. 2025, 09:50 → 10:50 Europe/Paris

Amphi Schwartz

A dynamical rare event is considered, e.g. using a Stochastic Differential Equation with small noise conditioned to hit set B before an attractor set A. Such rare events can be sampled using a population Monte Carlo method of splitting type (named Adaptive Multilevel Splitting) which sequentially kills and splits clones in a sample using a score function that evaluates proximity to B. We will review recent mathematical results on the small noise fluctuations of such algorithms with focus on the critical role of the score function in practice.