江苏省应用数学(中国矿业大学)中心系列学术报告
报告题目:Neural networks and solving high-dimensuional PDEs
报告人: 刘祖汉 教授 单位: 扬州大学数学学院
报告时间:2026年9月19日(周六)上午 9:30-10:30
报告地点:数学学院A321
欢迎全校师生参加!
数学学院
报告人及报告内容简介:
刘祖汉, 教授, 2003年入选教育部“优秀青年教师资助计划”。2003年12月起,历任江苏师范大学党委常委、副校长,扬州大学党委常委、副校长、副书记;2018年10月至2022年8月任盐城工学院党委书记。长期从事偏微分方程研究,先后主持多项国家自然科学基金项目,在SIAM J. Math. Anal.,J. Funct. Anal.,SIAM J. Appl. Math., JDE,CVPDE,European J. Applied. Math., J. Nonlinear Sci.等重要国际数学期刊上发表研究论文100余篇。
Abstract: This talk is primarily about the idea that underlie deep learning. The first part of the talk introduces deep learning models and considers architacture that are specialized to image, text, and graph data. The next part tackles solving high-dimensionsl PDEs. We prove that the source of the equation lies the spectral Barron space and the potential function with a non-negative lower boun decomposes as a positive constant plus a function of Barron space, the solution lies in the spectral Barron space. Finally, we prove that the solution to this PDE can be approximated on any bounded domain by a two-layer neural network with respect to the H^1-norm without the curse of dimensionality.
