Generative Neural Reparameterization for Differentiable PDE-constrained Optimization
Fuente:
arXiv
Gespeichert in:
| 1. Verfasser: | |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916442084999168 |
|---|---|
| author | Joglekar, Archis S. |
| author_facet | Joglekar, Archis S. |
| contents | Partial-differential-equation (PDE)-constrained optimization is a well-worn technique for acquiring optimal parameters of systems governed by PDEs. However, this approach is limited to providing a single set of optimal parameters per optimization. Given a differentiable PDE solver, if the free parameters are reparameterized as the output of a neural network, that neural network can be trained to learn a map from a probability distribution to the distribution of optimal parameters. This proves useful in the case where there are many well performing local minima for the PDE. We apply this technique to train a neural network that generates optimal parameters that minimize laser-plasma instabilities relevant to laser fusion and show that the neural network generates many well performing and diverse minima. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_12683 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Generative Neural Reparameterization for Differentiable PDE-constrained Optimization Joglekar, Archis S. Computational Physics Artificial Intelligence Machine Learning Numerical Analysis Plasma Physics Partial-differential-equation (PDE)-constrained optimization is a well-worn technique for acquiring optimal parameters of systems governed by PDEs. However, this approach is limited to providing a single set of optimal parameters per optimization. Given a differentiable PDE solver, if the free parameters are reparameterized as the output of a neural network, that neural network can be trained to learn a map from a probability distribution to the distribution of optimal parameters. This proves useful in the case where there are many well performing local minima for the PDE. We apply this technique to train a neural network that generates optimal parameters that minimize laser-plasma instabilities relevant to laser fusion and show that the neural network generates many well performing and diverse minima. |
| title | Generative Neural Reparameterization for Differentiable PDE-constrained Optimization |
| topic | Computational Physics Artificial Intelligence Machine Learning Numerical Analysis Plasma Physics |
| url | https://arxiv.org/abs/2410.12683 |