Reproduce “State-of-the-Art Review of Design of Experiments for Physics-Informed Deep Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sourav Das, Solomon Tesfamariam
A comprehensive review evaluates various Design of Experiments schemes for training Physics-Informed Neural Networks, demonstrating that quasi-random sampling methods, particularly Hammersley sampling, consistently yield more accurate predictions across different PDEs compared to classical approaches.
Key insights
- Quasi-random sampling methods generally offer superior space-filling properties, which are critical for the accuracy and performance of Physics-Informed Neural Networks (PINNs).
- Increasing the number of sampling points typically improves PINN accuracy, but with diminishing returns beyond a certain threshold.
- The relative performance of different Design of Experiments (DoE) methods can vary based on specific PDE characteristics, though quasi-random methods generally maintain an advantage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.