报告题目：Neural learning of Koopman operators

报告人：张宏坤教授，大湾区大学

报告时间：2025年11月17日（周一）10:00

报告地点：20-306

报告摘要：We address the inverse problem of learning invariant structures directly from trajectory data.Using symbolic dynamics, we construct dictionaries that align with the system’s generating partitions, yielding Koopman operator approximations naturally consistent with the underlying dynamics.Through Perron–Frobenius duality, the symbolic EDMD framework simultaneously recovers the Koopman spectrum and the invariant measure—specifically, the measure of maximal entropy (MME).Applications to nonlinear and intermittent maps demonstrate that both spectral and measure-theoretic features can be learned directly from orbit data, bridging data-driven Koopman analysis with classical ergodic theory.

报告人简介：张宏坤，大湾区大学讲席教授，原美国麻省大学终身教授。长期从事混沌动力系统、哈密顿动力系统与机器学习的交叉研究，系统提出多种动力系统可解释建模与学习方法，在动力学与深度学习双向赋能方面做出开创性贡献，推动人工智能与动力系统的深度融合与理论创新。已发表高水平学术论文70余篇，Google Scholar引用近千次。曾获美国国家科学基金会青年奖（NSF CAREER Award）、西蒙斯学者奖（Simons Fellow）及澳大利亚 Ether Raybould 访问学者奖，并多次担任国际学术会议主要组织者。

邀请人：动力系统与非线性分析研究所