Research map / Architectures and efficiency / Attention and transformers
CoEvol-NO: State and Coordinate Co-Evolution with an Error-Driven Predictor-Corrector Paradigm for Neural Operator Transformer
Jianqiao Zeng, Ruocheng Wang, Yanzhi Liu, Hao Xiong, 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 |
|---|---|---|---|
| update the mesh sequence jointly with the latent state (co-evolution) instead of evolving state only (ablation) | CoEvol-NO · Elasticity | relative L2 error: 0.3278 -> 0.0038 | Improved |
| keep a persistent latent state instead of recomputing state from mesh each layer (ablation) | CoEvol-NO · Navier-Stokes | relative L2 error: 0.1200 -> 0.0731 | Improved |
| use exact gradient correction instead of first-order approximation (ablation) | CoEvol-NO-dp · Navier-Stokes | relative L2 error: 0.0830 -> 0.0731 | Improved |
| add PC correction to Transolver attention and FFN | Transolver · Darcy | relative L2 error: 0.0057 -> 0.0043 | Improved |
| add PC correction to LNO self-attention | LNO · Airfoil | relative L2 error: 0.0196 -> 0.0082 | Improved |
| add PC correction to LNO self-attention | LNO · Pipe | relative L2 error: 0.0131 -> 0.0141 | Hurt |
| add layer-wise momentum on the latent state (beta_S=0.9) (ablation) | CoEvol-NO · Pipe | relative L2 error: 0.0044 -> 0.0032 | Improved |
| add momentum on the mesh sequence X (beta_X) (ablation) | CoEvol-NO · Pipe, Airfoil, Darcy | relative L2 error: Pipe 0.0044 -> 0.0048 at beta_X=0.9 | No clear effect |
| use learnable LayerScale step sizes instead of fixed scale (ablation) | CoEvol-NO · Pipe | relative L2 error: 0.0045 -> 0.0032 | Improved |
| add LayerScale to Transolver | Transolver · Airfoil | relative L2 error: 0.0057 -> 0.011 | Hurt |
| apply PC to all modules with exact gradient (ablation) | CoEvol-NO · Pipe | relative L2 error: 0.0033 -> 0.0049 | Hurt |
| remove PC from latent-only CoEvol-NO (ablation) | CoEvol-NO-Latent · Airfoil | relative L2 error: 0.0223 -> 0.0069 | Improved |
Closest papers by meaning
- MaRK: Markov-adapted Recurrent Kernels for Dynamic Operator Conditioning in State Space Models (NeurIPS 2026 · similarity 0.87)
- DISCO: learning to DISCover an evolution Operator for multi-physics-agnostic prediction (ICML 2025 · similarity 0.86)
- Adaptive Coordinate Transforms for Neural Operators (NeurIPS 2026 · similarity 0.86)
- Learning semilinear neural operators: A unified recursive framework for prediction and data assimilation. (ICLR 2024 · similarity 0.85)
- PhysGuard: Fisher-Guided Gradient Projection for Sim-to-Real Neural PDE Surrogates (NeurIPS 2026 · similarity 0.85)
- Transformers Converge to Invariant Algorithmic Cores (NeurIPS 2026 · 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.
