Research map / Probabilistic and scientific modeling / Differential equations and PDEs
ComPhy: Composing Physical Models with end-to-end Alignment
Alessandro Trenta, Andrea Cossu, Davide Bacciu
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 derivative-based alignment (DERL) instead of output-only alignment (OUTL) (ablation) | CP-2xPINN · Kovasznay flow | L2 error (x1e-6): 20.47 -> 7.221 | Improved |
| use derivative-based alignment (DERL) instead of output-only alignment (OUTL) (ablation) | CP-3xPINN · Acoustics | L2 error (x1e-5): 11.43 -> 7.449 | Improved |
| use derivative-based alignment (DERL) instead of output-only alignment (OUTL) (ablation) | CP-PINN+NCL · Taylor-Green vortex | L2 error (x1e-5): 6.024 -> 3.247 | Improved |
| add gradient-based reweighting to CP (ablation) | CP-3xNCL (DERL) · Acoustics | L2 error (x1e-5): 5.165 -> 2.718 | Improved |
| add gradient-based reweighting to CP (ablation) | CP-PINN+NCL (DERL) · Taylor-Green vortex | L2 error (x1e-5): 3.247 -> 2.795 | Improved |
| add gradient-based reweighting to a single PINN | PINN · NS-Euler gas | L2 error (x1e-3): 1.712 -> 3.319 | Hurt |
| replace single PINN with ComPhy modular alignment of several PINN modules | PINN · MHD | L2 error (x1e-4): 1.967 -> 1.535 | Improved |
| replace single PINN with ComPhy CP-2xPINN (SOB) | PINN · NS-Euler gas | L2 error (x1e-3): 1.712 -> 1.296 | Improved |
| use SOB alignment with two NCL modules (ablation) | CP-2xNCL · NS-Euler gas | L2 error (x1e-3): 1.396 -> 2.029 | Hurt |
| lower lambda_align from 1 to 0.01 (ablation) | CP-3xNCL (DERL) · Acoustics | L2 error (x1e-5): 5.165 -> 2.460 | Improved |
| raise lambda_align from 1 to 10 (ablation) | CP-3xNCL (DERL) · Acoustics | L2 error (x1e-5): 5.165 -> 12.46 | Hurt |
| compare CP-3xNCL with a 3x larger NCL at matched training parameters (ablation) | CP-3xNCL (DERL+Grad) · Acoustics | L2 error (x1e-5): 2.940 -> 2.718 | Improved |
Closest papers by meaning
- Holistic Physics Solver: Learning PDEs in a Unified Spectral-Physical Space (ICML 2025 · similarity 0.86)
- AROMA: Preserving Spatial Structure for Latent PDE Modeling with Local Neural Fields (NeurIPS 2024 · similarity 0.86)
- Text2PDE: Latent Diffusion Models for Accessible Physics Simulation (ICLR 2025 · similarity 0.86)
- Physics-informed Reduced Order Modeling of Time-dependent PDEs via Differentiable Solvers (NeurIPS 2025 · similarity 0.85)
- P$^2$C$^2$Net: PDE-Preserved Coarse Correction Network for efficient prediction of spatiotemporal dynamics (NeurIPS 2024 · similarity 0.85)
- Poseidon: Efficient Foundation Models for PDEs (NeurIPS 2024 · 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.
