This is absolutely amazing! I never thought to use finite difference methods inside the loss function to compute derivatives. I tried to build PINNs in WL in the past , but I gave up after realizing that WL’s neural network library based on mxnet could not automatically compute higher-order derivatives like TensorFlow or PyTorch. Your approach appears to be a promising path forward.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| PINN: Physics Informed Neural Networks for Laplace PDE on L-shaped domain | 6 | 14567 | December 5, 2023 | |
| Solving PDEs using neural networks? | 5 | 12800 | July 2, 2021 | |
| A machine learning framework based on physical information: a little problem! | 4 | 3776 | May 10, 2024 | |
| [WSS21] Use neural network to predict the solution of partial differential | 2 | 8352 | May 6, 2024 | |
| [WSS18] Teaching neural networks to solve partial differential equations | 2 | 14542 | October 20, 2023 |