ABSTRACT.
Many problems in sampling and machine learning can be viewed as the evolution of large systems of interacting particles, representing samples, agents, or model parameters. At the microscopic level, their dynamics are described by ordinary or stochastic differential equations; at the macroscopic level, the corresponding probability distribution evolves according to a nonlinear PDE. This mean-field viewpoint connects probability, PDEs, optimal transport, particle systems, and machine learning. A central challenge is to understand their long-time behavior: Do they converge to a desired distribution? Which equilibria are stable, and when do instability or pattern formation occur? I will introduce this perspective through examples from sampling, aggregation equations, and mean-field learning dynamics, emphasizing the main ideas and sone open questions.