Speaker: Liu Jun, Harvard University

Time: 15:00-16:00, March 7, 2025, GMT+8

Venue: Sanyuan Hall, Zhihua Building

Abstract: Particle filtering, also known as sequential Monte Carlo, has been widely used as a set of powerful computational tools for making Bayesian and likelihood-based inference in both static and dynamical systems.  We will review its  historical developments and recent advances, discussing a few strategies  that may be useful for dealing with complex problems. Particle filtering is built upon the ideas of importance sampling with resampling, in which resampling plays a critical, yet mysterious (still), role of steering the algorithm’s intermediate outcomes towards the “future” (or goal). Multinomial resampling and its variations have been most widely adopted in the past. Some recent ideas include stratified resampling, optimal-transport resampling, gradient-based moves, etc. We will discuss these ideas and their extensions. This talk is based on the joint work with Chunlin Ji, Wenshuo Wang, Yichao Li, and Ke Deng.

Source: School of Mathematical Sciences