Deep Learning Method for Plasma Simulation

Description

Awards:
1. School Winner
2. Student-Nominated Winner
3. Excellent Poster

Abstract: In this project, we propose a gradient-based adaptive sampling strategy for the deep unfitted Nitsche method to solve the elliptic interface problem. The deep unfitted Nitsche method involves training two weakly coupled neural networks by minimizing an energy formualtion to address the discontinuities caused by the interface. The gradient-based adaptive sampling strategy is then designed to capture the singularity at the interface for more challenging problems. The algorithm’s performance is presented through a numerical example to demonstrate its computational efficiency and accuracy.

Period	17 Jun 2024 → 26 Aug 2024