Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Domain Agnostic Fourier Neural Operators
Ning Liu, Siavash Jafarzadeh, Yue Yu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add smoothed boundary encoding to FNO rather than using a sharp mask (ablation) | FNO · Hyperelasticity, 1000 training samples | relative L2 error: 4.550%$\pm$0.062% -> 1.415\%$\pm$0.025\% | Improved |
| Add smoothed boundary encoding to IFNO rather than using a sharp mask (ablation) | IFNO · Hyperelasticity, 1000 training samples | relative L2 error: 4.481%$\pm$0.022% -> 1.509\%$\pm$0.018\% | Improved |
| Encode geometry in the iterative Fourier layer rather than provide smoothed χ only as input (ablation) | eDAFNO versus FNO w/ smooth χ · Hyperelasticity, 1000 training samples | relative L2 error: 1.415\%$\pm$0.025\% -> 1.094%$\pm$0.012% | Improved |
| Encode geometry in the iterative Fourier layer rather than provide smoothed χ only as input (ablation) | iDAFNO versus IFNO w/ smooth χ · Hyperelasticity, 1000 training samples | relative L2 error: 1.509\%$\pm$0.018\% -> 1.207%$\pm$0.006% | Improved |
| Use layer-independent parameters instead of layer-specific parameters (ablation) | DAFNO · Hyperelasticity | relative L2 error: 1.094%$\pm$\bf{0.012}% vs 1.207%$\pm$0.006%; 4.247\%$\pm$0.066\% vs \bf{4.214}\%$\pm$\bf{0.058}\% | Mixed |
| Use irregular rather than uniform grids (ablation) | eDAFNO · Airfoil test | relative L2 error: 0.596\%$\pm$\bf{0.005}\% -> 0.659\%$\pm$0.007\% | Hurt |
| Tune smoothing coefficient β across the tested range (ablation) | eDAFNO · Hyperelasticity, 1000 training samples | test loss: 1.155% (β=5); 1.064% (β=10); 1.120% (β=20); 1.147% (β=30); 1.135% (β=40); 1.179% (β=50); 1.194% (β=60); 1.240% (β=70); 1 | Mixed |
| Use eDAFNO instead of Geo-FNO | Airfoil design · Airfoil test | relative L2 error: 1.650\%$\pm$0.175\% -> \bf{0.596}\%$\pm$\bf{0.005}\% | Improved |
| Train eDAFNO on irregular grids | Airfoil design · Airfoil test | relative L2 error: Geo-FNO 1.650\%$\pm$0.175\%; eDAFNO on irregular grids 0.659\%$\pm$0.007\% | Improved |
| Apply eDAFNO to pipe flow rather than Geo-FNO | Pipe flow · Pipe flow test | relative L2 error: 0.67\% -> 0.71\% | Hurt |
| Apply eDAFNO to pipe flow rather than Geo-FNO | Pipe flow · Pipe flow test | maximum absolute error: 0.061 -> 0.051 | Improved |
Closest papers by meaning
- Beyond Regular Grids: Fourier-Based Neural Operators on Arbitrary Domains (ICML 2024 · similarity 0.88)
- Component Fourier Neural Operator for Singularly Perturbed Differential Equations (AAAI 2024 · similarity 0.87)
- Amortized Fourier Neural Operators (NeurIPS 2024 · similarity 0.87)
- Guaranteed Approximation Bounds for Mixed-Precision Neural Operators (ICLR 2024 · similarity 0.86)
- Neural Operators with Localized Integral and Differential Kernels (ICML 2024 · similarity 0.85)
- Reference Neural Operators: Learning the Smooth Dependence of Solutions of PDEs on Geometric Deformations (ICML 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
