Research map / Graph learning / GNN architectures
Smoothness Errors in Dynamics Models and How to Avoid Them
Edward Berman, Luisa Li, Jung Yeon Park, Robin Walters
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 Taylor-truncated relaxed unitary convolution instead of GCN | R-UniGraph · Heat diffusion on grid graphs | MSE: 1.08 · 10−2 -> 0.11 · 10−2 | Improved |
| Relax unitary constraint (Taylor truncation) instead of strict Lie unitary (ablation) | R-UniGraph · Heat diffusion on grid graphs | mean Rayleigh quotient error: 8.86 · 10−2 -> 2.07 · 10−2 | Improved |
| Use R-UniMesh instead of EMAN and Hermes | R-UniMesh · Heat MeshPDE | NRMSE: 51.9 vs 73.50 (EMAN) and 73.02 (Hermes) | Improved |
| Use R-UniMesh | R-UniMesh · Wave MeshPDE | NRMSE and Rayleigh error: NRMSE 236.5 best; RE 93.5 vs Transformer 48.0 | Mixed |
| Use R-UniMesh | R-UniMesh · Cahn-Hilliard MeshPDE | NRMSE: 123.9 vs GemCNN 121.2 | No clear effect |
| Increase Taylor truncation Tmax (ablation) | R-UniGraph · Heat diffusion | MSE | Hurt |
| Use Tmax = 1 (GCN-like with hermitian weights) (ablation) | R-UniGraph · Heat diffusion | smoothness and MSE | Mixed |
| Replace unitary convolution layers with ordinary graph convolutions (ablation) | R-UniMesh · Diffusive dynamics | accuracy | Hurt |
| Add geometric inductive biases (unitarity or gauge equivariance) | Mesh GNNs · Unseen complex meshes | accuracy | Improved |
| Use unitary or equivariant models on WeatherBench2 | R-UniMesh and equivariant models · WeatherBench2 | RMSE | No clear effect |
Closest papers by meaning
- CALM-PDE: Continuous and Adaptive Convolutions for Latent Space Modeling of Time-dependent PDEs (NeurIPS 2025 · similarity 0.87)
- Mesh-Free Convolution: Learned Spectral Attenuation and Transport (NeurIPS 2026 · similarity 0.87)
- Hyperbolic-PDE GNN: Spectral Graph Neural Networks in the Perspective of A System of Hyperbolic Partial Differential Equations (ICML 2025 · similarity 0.87)
- Unitary Convolutions for Learning on Graphs and Groups (NeurIPS 2024 · similarity 0.87)
- Graph Neural PDE Solvers with Conservation and Similarity-Equivariance (ICML 2024 · similarity 0.86)
- P$^2$C$^2$Net: PDE-Preserved Coarse Correction Network for efficient prediction of spatiotemporal dynamics (NeurIPS 2024 · similarity 0.86)
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.
