系列报告

In many scientific fields, from molecular dynamics to single-cell genomics, we encounter a common data challenge: high-dimensional observations are collected, but their temporal labels are missing. This transforms sequential trajectory data into an unordered point cloud, breaking the standard paradigm for dynamical system identification. This talk presents two complementary frameworks for reconstructing the underlying dynamics and recovering the lost time labels. We begin with a method that treats the data as samples from a probability distribution. By formulating the recovery task as a distribution-matching problem, specifically minimizing the Sliced Wasserstein Distance between observed and simulated data.

史作强，清华大学丘成桐数学科学中心长聘教授、博士生导师，同时担任北京雁栖湖应用数学研究院兼职研究员。他在清华大学完成本科与博士阶段学习，曾赴美国加州理工学院从事博士后研究，在美国加州大学洛杉矶分校数学系担任访问学者。他主持过国家自然科学基金相关项目，在ACHA、SIAM系列期刊等国际知名学术刊物发表论文50余篇，研究方向覆盖偏微分方程数值方法、图像处理与机器学习中的微分方程模型、非线性非平稳信号时频分析等领域。