Research map / Architectures and efficiency / Attention and transformers
Hyperbolic Neural Operator
Jieyuan Pei, Zhuoxuan Li, Wei Li, Haobo Zhang, jiawei jiang, Jianwei Zheng
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 hyperbolic distance logits instead of Euclidean logits (ablation) | HNO · Darcy | relative L2 error: 0.00446 -> 0.00493 (+10.5%) when removed | Improved |
| add adaptive temperature instead of fixed tau (ablation) | HNO · Darcy | relative L2 error: 0.00446 -> 0.00464 (+4.0%) when removed | Improved |
| use hyperbolic distance instead of Euclidean dot-product routing (ablation) | HNO · Elasticity | relative L2 error (x1e-3): 3.688 -> 4.100 when removed | Improved |
| use hyperbolic distance instead of Euclidean dot-product routing (ablation) | HNO · Airfoil volume | relative L2 error (x1e-3): 12.009 -> 16.683 when removed | Improved |
| use hyperbolic radius instead of Euclidean learned temperature gate (ablation) | HNO · Elasticity | relative L2 error (x1e-3): 3.688 -> 5.422 when replaced by TempGate | Improved |
| use hyperbolic radius instead of Euclidean RBF width gate (ablation) | HNO · Airfoil volume | relative L2 error (x1e-3): 12.009 -> 133.168 when replaced by RBFGate | Improved |
| keep learned Q/K norms rather than fixing them at inference (ablation) | HNO · Darcy | relative L2 error: 0.00446 -> 0.03269 when norm fixed | Improved |
| keep sample-specific Q/K norms rather than shuffling them (ablation) | HNO · Darcy | relative L2 error: 0.00446 -> 0.03328 when norm shuffled | Improved |
| apply hyperbolic geometry only to attention logits rather than the full manifold-valued backbone (ablation) | HNO · Darcy | relative L2 error: 0.00446 -> 0.00817 with full manifold backbone | Improved |
| use token-mediated compression instead of point-to-grid patchification (ablation) | HNO · Darcy | relative L2 error: 0.00446 -> 0.00744 with H-SparsePatch | Improved |
| use HNO instead of Transolver++ | Transolver++ · Elasticity | relative L2 error: 0.0064 -> 0.0037 | Improved |
| use HNO instead of Transolver | Transolver · Darcy | relative L2 error and params: 0.00567 -> 0.00446; 2.83M -> 0.82M | Improved |
Closest papers by meaning
- RIGNO: A Graph-based Framework For Robust And Accurate Operator Learning For PDEs On Arbitrary Domains (NeurIPS 2025 · similarity 0.88)
- HyPINO: Multi-Physics Neural Operators via HyperPINNs and the Method of Manufactured Solutions (NeurIPS 2025 · similarity 0.88)
- Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators (NeurIPS 2024 · similarity 0.87)
- Geometry Aware Operator Transformer as an efficient and accurate neural surrogate for PDEs on arbitrary domains (NeurIPS 2025 · similarity 0.86)
- HEAP: Hyper Extended A-PDHG Operator for Constrained High-dim PDEs (ICML 2025 · similarity 0.86)
- Latent Neural Operator for Solving Forward and Inverse PDE Problems (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.
