I see. So this is a limitation of Zygote + ForwardDiff specifically? I’ll try the implementation with TaylorDiff next. I’m aware that Lux is better behaved than Flux for a variety of applications, but it shouldn’t make a difference in this case, correct? (Given that Lux is also built on top of Zygote).
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Higher order derivatives/ automatic differentiation | 5 | 743 | December 26, 2023 | |
| What is the best way to implement PINN in Julia | 18 | 757 | April 6, 2025 | |
| Nested and different AD methods altogether: How to add AD calculations inside my loss function when using neural differential equations? | 9 | 1046 | September 28, 2024 | |
| Flux PINN 1D Burgers | 11 | 923 | May 23, 2025 | |
| Nested AD with Lux etc | 26 | 1412 | May 1, 2024 |