Research map / Probabilistic and scientific modeling / Differential equations and PDEs
MAgNet: Mesh Agnostic Neural PDE Solver
Oussama Boussif, Yoshua Bengio, Loubna Benabbou, Dan Assouline
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 |
|---|---|---|---|
| Use MAgNet[CNN] instead of FNO or MPNN on E1 irregular meshes | E1 solver · E1 irregular meshes, varied train/test resolutions | MAE | Improved |
| Use MAgNet[GNN] instead of MPNN on E1 irregular meshes | E1 solver · E1 irregular meshes, varied train/test resolutions | MAE | Mixed |
| Add graph interactions between spatial queries and parent-mesh coordinates during forecasting (ablation) | MAgNet[CNN] · E1 regular mesh, train nx=50; test nx'=40 | MAE: 0.1650 -> 0.1079 | Improved |
| Add graph interactions between spatial queries and parent-mesh coordinates during forecasting (ablation) | MAgNet[CNN] · E1 regular mesh, train nx=50; test nx'=50 | MAE: 0.0815 -> 0.1020 | Hurt |
| Add graph interactions between spatial queries and parent-mesh coordinates during forecasting (ablation) | MAgNet[CNN] · E1 regular mesh, train nx=50; test nx'=100 | MAE: 0.2810 -> 0.1142 | Improved |
| Add graph interactions between spatial queries and parent-mesh coordinates during forecasting (ablation) | MAgNet[CNN] · E1 regular mesh, train nx=50; test nx'=200 | MAE: 0.4139 -> 0.1177 | Improved |
| Replace the learned interpolator with a basic cubic interpolator (ablation) | MAgNet[CNN] · E1 regular meshes, test nx' in {40,50,100,200} | MAE | No clear effect |
| Increase the number of spatial queries sampled during training (ablation) | MAgNet[GNN] · E1, uniform mesh, nx=50 | MAE | Improved |
| Replace learned interpolation with KNN, linear, or cubic interpolation on original frames (ablation) | MAgNet[CNN] · E1 regular meshes, nx=50 | MAE | No clear effect |
Closest papers by meaning
- Graph Neural PDE Solvers with Conservation and Similarity-Equivariance (ICML 2024 · similarity 0.90)
- P$^2$C$^2$Net: PDE-Preserved Coarse Correction Network for efficient prediction of spatiotemporal dynamics (NeurIPS 2024 · similarity 0.89)
- Better Neural PDE Solvers Through Data-Free Mesh Movers (ICLR 2024 · similarity 0.88)
- PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systems (ICLR 2025 · similarity 0.88)
- RIGNO: A Graph-based Framework For Robust And Accurate Operator Learning For PDEs On Arbitrary Domains (NeurIPS 2025 · similarity 0.88)
- Geometric and Physical Constraints Synergistically Enhance Neural PDE Surrogates (ICML 2025 · similarity 0.87)
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.
