Research map / Architectures and efficiency / Attention and transformers
ReViT: Rotational-equivariant Vision Transformers for Neural PDE Solvers
Hao Wei, Björn List, Nils Thuerey
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 |
|---|---|---|---|
| replace invariant embedding with standard CNN tokenization (ablation) | ReViT · 3D equivariance (stage-wise R2) | R2 correlation | Hurt |
| use rebased relative positional encodings instead of standard relative PEs (ablation) | ReViT · 3D equivariance (stage-wise R2) | R2 correlation | Improved |
| use equivariant decoder instead of standard CNN reconstruction head (ablation) | ReViT · 3D equivariance (output R2) | R2 correlation | Improved |
| add hard-coded rotational equivariance (ReViT) over non-equivariant PDE-Transformer | PDE-Transformer · 2D Advection | MSE: MSE reduction around 37.2% | Improved |
| add rotated training data (+pi/3) to PDETrans (ablation) | PDE-Transformer · 2D Kolmogorov Flow | MSE: 0.2125 -> 0.2256 | No clear effect |
| add rotated training data (+pi/3, +pi/6) to ReViT (ablation) | ReViT · 2D Kolmogorov Flow | MSE: 0.1361 -> 0.0803 | Improved |
| add random angular augmentation to PDETrans (ablation) | PDE-Transformer · 2D Kolmogorov Flow | MSE: 0.2125 -> 0.1853 | Improved |
| use ReViT instead of AViT on 3D MHD | 3D MHD · 3D MHD | MSE: 2.20 -> 0.82 (x10^-2) | Improved |
| use ReViT on wall-bounded anisotropic flow | 3D TCF · 3D TCF | MSE: 0.60 -> 0.21 (x10^-2) | Improved |
| replace lifted group attention with canonicalization-based ReViT | GE-ViT / GSA-Nets · RotMNIST | memory: 95.5 GB -> 1.81 GB | Improved |
| add intra-patch positional encoding (ablation) | ReViT · ablation (Appendix J.3) | performance | No clear effect |
| use analytical AOV basis instead of eigendecomposition-based BOA basis (ablation) | ReViT · basis construction latency, 2D res 32 | latency (ms): 1.47 -> 0.65 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AViT — A-ViT: Adaptive Tokens for Efficient Vision Transformer (CVPR 2022) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EqGINO: Equivariant Geometry-Informed Fourier Neural Operators for 3D PDEs (ICML 2026 · similarity 0.88)
- PDE-Transformer: Efficient and Versatile Transformers for Physics Simulations (ICML 2025 · similarity 0.87)
- Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators (NeurIPS 2024 · similarity 0.87)
- Lie Algebra Canonicalization: Equivariant Neural Operators under arbitrary Lie Groups (ICLR 2025 · similarity 0.87)
- CViT: Continuous Vision Transformer for Operator Learning (ICLR 2025 · similarity 0.86)
- CALM-PDE: Continuous and Adaptive Convolutions for Latent Space Modeling of Time-dependent PDEs (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.
