Petra Lazić: Interacting particle systems as tools in optimization

Abstract: Arising from applications in machine learning such as the problem of approximate sampling from a given probability distribution or optimization, special types of stochastic processes such as interacting particle systems became highly attractive in recent years. The theory about them is quite interesting as it shows that in many ways they exhibit different characteristics than classical diffusion processes. Furthermore, there is a close relationship between them and certain evolution PDEs, enabling the model to be viewed from various angles. In this talk, I will discuss the behaviour of interacting particle systems and their gradient flows, focusing on the applied aspects. In particular, I will present numerical methods that can be used for simulations of these type of processes and discuss various optimization problems for which they come in handy as a tool.

The seminar now has a homepage with the planned talks for the entire semester: https://mathematical-conversations.splet.arnes.si/