|
Finite-element-based physics-informed neural networks (FE-PINNs) provide a strategy for training computationally efficient and flexible surrogate models. Existing FE-PINN architectures only feature custom convolutional operators, which greatly limits architecture design. In this project, the student will implement new custom operators that enable encoder-decoder style networks (e.g., U-Net). These custom operators will be incorporated into the existing software package for FE-PINN training.
|
Sign in
to view more information about this project.
