报告人:张志跃 教授 南京师范大学
报告题目:Machine learning for interface problems and some applications
摘 要:Three accurate and efficient numerical methods have been proposed for interface problems with extreme conditions such as very big jump ratio, coefficient blow-up and geometric singularity interface. One of the schemes combines DNN technique with FDM to build decoupling second order method for the problem. The second one based on operator learning to construct predictor corrector method for interface problems, the main advantages of this method is saving computational costs and having higher accuracy, in addition, it can be applied to moving interface and 3D interface problems. The last one proposed a perturbation-correction framework based on Local Randomized Neural Networks(LRaNNs) for nonlinear interface problem. Examples confirmed theoretical results. Recent numerical results of some interesting problems for shocks and free interface have been shown in this talk.
报告人简介:张志跃,南京师范大学数学科学学院教授,博士生导师。2001年毕业于山东大学数学与系统科学学院,2002年至2004年曾在中国科学院大气物理研究所进行博士后研究工作。在第八届世界华人数学家大会做45分钟邀请报告,现任美国《数学评论》评论员、中国工业与应用数学学会油水资源专业委员会成员。对退化问题提出了混合增广高精度数值方法。主持完成多项国家自然科学基金面上项目和江苏省自然科学基金面上项目,发表学术论文八十余篇。目前主要的研究兴趣为偏微分方程数值解、PDE约束最优控制问题、机器学习及计算流体力学。
报告时间:2026年6月12日,15:00--16:30
报告地点:玉泉校区创新北楼512会议室
欢迎广大师生参加!
内蒙古大学数学科学学院
2026年6月3日