报告题目 (Title)：A novel deep convolutional surrogate model with incomplete solve loss for parameterized steady-state diffusion problems（参数化稳态扩散问题的一种新的具有不完全解损失的深度卷积代理模型）
报告人 (Speaker)：张晓平 副教授（武汉大学）
报告时间 (Time)：2025年7月13日（周日）9:30
报告地点 (Place)：校本部GJ406
邀请人(Inviter)：刘东杰
主办部门：理学院数学系
报告摘要： In this talk, we will introduce a novel deep surrogate model that integrates the generalization capabilities of convolutional neural networks (CNNs) with traditional numerical methods to solve parametrized steady-state diffusion problems. We will adopt different strategies to handle linear and nonlinear cases separately. In order to solve linear problems, a novel loss function is designed based on an iterative solver for unsupervised training of the model. To solve nonlinear problems, Picard iterations are integrated into the training strategy for unsupervised model training. Extensive numerical experiments are used to valid our method and massive numerical results have shown that our deep surrogate method is capable to solve various parametrized diffusion problems.
