- 工程数学讨论班：控制、滤波、状态估计与信号处理的基础与前沿

Nonlinear Filters and Neural Networks with Recursive Structure

组织者

焦小沛 , 康家熠

演讲者

孙泽钜

时间

2025年04月10日 14:30 至 16:00

地点

A3-2-303

线上

Zoom 435 529 7909 (BIMSA)

摘要

Filter is a subject about the estimation of a stochastic dynamical system based on a sequential of noisy observations. When it comes to systems with high-dimensionality and high-nonlinearity, the performance of classical filtering algorithms is often unsatisfactory, either in accuracy or in computational efficiency. With the development of artificial intelligence (AI), whether and how AI can solve nonlinear filtering problems has become a hotspot issue for both researchers and practitioners. In this talk, I will present the relationship between nonlinear filters and neural networks with recursive structure, and discuss the capability of this kind of neural networks to solve nonlinear filtering problems, especially in the case of systems with high-dimensionality and high-nonlinearity. The contents of this talk are based on the joint work with Dr. Xiuqiong Chen, Dr. Yangtianze Tao and Prof. Stephen S.-T. Yau.