北京雁栖湖应用数学研究院

丘成栋

张晓明

2024年07月03日 14:45 至 15:45

理科楼A-304

Yau-Yau filter is the first method potentially suitable for applications to practically any nonlinear filtering problem. This method converts solving a nonlinear filtering problem into solving a linear Forward Kolmogorov Equation (FKE) computationally. Using accuracy and efficiency as two important criteria for developing and evaluating different implementation methods, we will discuss the three difficulties associated with the methods: nonlinearity and dimensionality of the state process and the observation process, and the explicit time-dependency in FKE. Tested with three illustrative cases, preliminary results suggest that PINN（Physics-informed Neural Network) may be one of the viable candidates for implementation.

张晓明博士先后在浙江大学、北京大学、麻省理工学院获得学士、硕士和博士学位。现任北京雁栖湖应用数学研究院研究员，人工智能和机器学习团队PI。他目前的研究兴趣是对开发由数据和领域知识驱动的机器学习算法，并将其应用于各种物理、生物和社会现象的解释和量化。