Sequential Monte Carlo methods with non-standard applications
数学专题报告
报告题目(Title)：
Sequential Monte Carlo methods with non-standard applications

报告人(Speaker)：
王亮亮（Simon Fraser University）

地点(Place)：
Zoom会议（会议号85839679693，密码406630）

时间(Time)：
2026年4月29日（周三）10:00-11:00

邀请人(Inviter)：
熊云丰
报告摘要
Modern Bayesian inference frequently requires sampling from complex, high-dimensional posterior distributions. While standard Markov chain Monte Carlo (MCMC) algorithms are foundational, they often exhibit poor mixing, struggle with local transition efficiency, and become trapped in local optima when applied to highly structured parameter spaces, such as discrete combinatorial spaces or multimodal posterior landscapes. In this talk, I will introduce Sequential Monte Carlo (SMC) methods—specifically Annealed Sequential Monte Carlo (ASMC)—as a powerful and "embarrassingly parallel" alternative for Bayesian computation. ASMC efficiently provides an approximate posterior distribution alongside an unbiased estimator for the marginal likelihood, which is highly valuable for conducting Bayesian model comparison and testing the correctness of posterior simulations. I will demonstrate the adaptability, scalability, and efficiency of SMC through two primary non-standard applications involving complex statistical models. First, I will discuss Bayesian phylogenetic inference, where we utilize SMC to navigate the highly discrete and computationally challenging space of evolutionary trees using biological sequence data. Second, I will present the application of SMC for parameter estimation and model selection in nonlinear ordinary differential equations (ODEs), with a specific focus on infectious disease transmission
--
FROM 202.120.11.*