Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Origo: Interpretable Multi-physics PDE Foundation Model through Neural Operator Splitting
Li Sun, Hongbo Lv, Zhikai Jiang, Zhongtian Sun, Lanxu Yang, Philip S. Yu
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 symmetric Strang split-stepper with a standard residual backbone (ablation) | Origo · NS | relative L2 error: 0.267 -> 0.0871 | Improved |
| add Strang split-stepper (ablation) | Origo · zero-shot 1D KdV | relative L2 error: 0.0421 -> 0.0103 | Improved |
| add Strang split-stepper (ablation) | Origo · zero-shot 2D Allen-Cahn | relative L2 error: 0.285 -> 0.0952 | Improved |
| add sparse mechanism inference routing (ablation) | Origo · Burgers | relative L2 error: 0.0231 -> 0.00989 | Improved |
| add sparse mechanism inference routing (ablation) | Origo · zero-shot 1D KdV | relative L2 error: 0.0284 -> 0.0103 | Improved |
| add sparse mechanism inference routing (ablation) | Origo · zero-shot 2D Allen-Cahn | relative L2 error: 0.193 -> 0.0952 | Improved |
| fine-tune Origo for 200 epochs | Origo-M · PDEArena NS | L2RE: 0.0871 -> 0.0247 | Improved |
| fine-tune Origo for 200 epochs | Origo-M · CFDBench | L2RE: 0.0125 -> 0.00443 | Improved |
| scale Origo from 20M to 260M parameters (ablation) | Origo · zero-shot | L2 relative error: 0.063 -> 0.032 | Improved |
| use Origo-M zero-shot versus second-best MoE-POT-M | MoE-POT-M · unseen KdV and Allen-Cahn | relative L2 error: 0.0237 -> 0.0103; 0.152 -> 0.0952 | Improved |
| Origo-M zero-shot versus MoE-POT-M on CFDBench | MoE-POT-M · CFDBench | L2RE: 0.00599 -> 0.0125 | Hurt |
Closest papers by meaning
- HyPINO: Multi-Physics Neural Operators via HyperPINNs and the Method of Manufactured Solutions (NeurIPS 2025 · similarity 0.88)
- Data-Efficient Operator Learning via Unsupervised Pretraining and In-Context Learning (NeurIPS 2024 · similarity 0.88)
- Holistic Physics Solver: Learning PDEs in a Unified Spectral-Physical Space (ICML 2025 · similarity 0.88)
- DISCO: learning to DISCover an evolution Operator for multi-physics-agnostic prediction (ICML 2025 · similarity 0.88)
- Pretraining Codomain Attention Neural Operators for Solving Multiphysics PDEs (NeurIPS 2024 · similarity 0.88)
- Neural Interpretable PDEs: Harmonizing Fourier Insights with Attention for Scalable and Interpretable Physics Discovery (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.
