报告承办单位: 数学与统计学院

报告内容: Transport Maps and Reinforcement Learning for Nonlinear Data Assimilation

报告人姓名: 姜立建

报告人所在单位: 同济大学

报告人职称/职务及学术头衔: 教授

报告时间:2026年10月28日 10:30

报告地点:长沙理工大学云塘校区理科楼A-212

邀请人：欧娜

报告摘要：Data assimilation is essential for state estimation and uncertainty quantification in dynamical systems. However, nonlinear dynamics, non-Gaussian distributions, and unknown governing models pose significant challenges. In this talk, I will present several approaches to data assimilation based on probability transport and reinforcement learning. First, we introduce an ensemble transport filter that constructs transport maps by minimizing maximum mean discrepancy (MMD) to approximate Bayesian posterior distributions. Next, we develop a coupling-informed transport filtering framework that exploits dependencies between state and observation variables for efficient, likelihood-free Bayesian updates. Finally, we present a reinforcement learning approach that learns unknown state transition models directly from noisy observations. These methods provide new perspectives and computational tools for nonlinear and high-dimensional data assimilation.

报告人简介：姜立建，同济大学数学院教授，主要研究方向是多尺度计算方法、模型降阶和不确定性量化。Texas A&M大学数学博士，曾在明尼苏达大学IMA和LANL国家实验室从事博士后研究和油服公司任首席科学家，国家级人才，曾主持多项国家自科基金项目，现任JCAM，JCMDS等多个期刊编委。