Research map / Architectures and efficiency / RNNs and normalization
G-RANS: Generalizable Residual-Aware Neural Solvers for Sparse Systems
Weixin Liao, Mingquan Feng, Zhizhou Zhang, Youjia Wu, Yifan Fu, Junchi Yan
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 multi-stage progressive bootstrap training (ablation) | G-RANS · Poisson n≈1000 ±10% | relative residual: 5.22e-4 -> 2.80e-5 | Improved |
| add multi-scale basis emission in the residual-aware basis generator (ablation) | G-RANS · Poisson n≈1000 ±10% | relative residual: 2.47e-2 -> 2.80e-5 | Improved |
| add both multi-scale and multi-stage (vs sub-NSS) (ablation) | G-RANS · Poisson n≈1000 ±10% | relative residual: 3.09e-2 -> 2.80e-5 | Improved |
| vary max active subspace dimension Kmax between 200 and 300 (ablation) | G-RANS · G-RANS | relative residual | No clear effect |
| use G-RANS instead of end-to-end GNO/FNO | G-RANS · n≈1000 FEM systems | relative residual: GNO/FNO ~1e-2 to 1e-1 -> G-RANS 1e-5 level | Improved |
| increase coefficient variation from ±10% to ±70% | G-RANS · Poisson n≈1000 | relative residual: 2.80e-05 -> 4.50e-05 | No clear effect |
| add Jacobi left preconditioning to G-RANS pipeline | G-RANS · Helmholtz and React.-Diff. | speedup over GMRES: 2.24x-2.26x with Jacobi vs about 3x without | Mixed |
| scale to 3D systems at n≈10^4 | G-RANS · Poisson-3D | speedup over GMRES: 2.36x (n≈1000) -> 1.14x (n≈10000) | Hurt |
| compare G-RANS wall-clock to GMRES/FGMRES at equal residual | G-RANS · n≈1000 FEM systems | speedup: 2-4x | Improved |
Closest papers by meaning
- UGrid: An Efficient-And-Rigorous Neural Multigrid Solver for Linear PDEs (ICML 2024 · similarity 0.86)
- MC^2: Monte Carlo Correction for Fast Elliptic PDE Solving (NeurIPS 2026 · similarity 0.86)
- Breakeven complexity: A new perspective on neural partial differential equation solvers (NeurIPS 2026 · similarity 0.85)
- RIGNO: A Graph-based Framework For Robust And Accurate Operator Learning For PDEs On Arbitrary Domains (NeurIPS 2025 · similarity 0.85)
- Transolver++: An Accurate Neural Solver for PDEs on Million-Scale Geometries (ICML 2025 · similarity 0.85)
- INC: An Indirect Neural Corrector for Auto-Regressive Hybrid PDE Solvers (NeurIPS 2025 · 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.
