Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Learning to Accelerate Partial Differential Equations via Latent Global Evolution
Tailin Wu, Takashi Maruyama, Jure Leskovec
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 global latent evolution instead of evolving states in input space (ablation) | LE-PDE on 3D Navier-Stokes · 3D Navier-Stokes flow, prediction at t=40 | runtime [s]: 1.03 -> 0.084 | Improved |
| Use latent-space boundary optimization | LE-PDE · 2D Navier-Stokes boundary inverse optimization | runtime [s]: 111.14 -> 49.81 | Improved |
| Use latent-space boundary optimization rather than input-space evolution (ablation) | LE-PDE · 2D Navier-Stokes boundary inverse optimization | GT-solver error: 0.305 -> 0.035 | Improved |
| Add multi-step loss (ablation) | LE-PDE · 1D nonlinear PDE E2, (250,50) | accumulated error: 3.337 -> 1.127 | Improved |
| Add multi-step loss (ablation) | LE-PDE · 2D Navier-Stokes, ν=10^-5, T=20, N=1000 | relative L2 error: 0.2156 -> 0.1861 | Improved |
| Add latent consistency loss (ablation) | LE-PDE · 1D nonlinear PDE E2, (250,50) | accumulated error: 6.386 -> 1.127 | Improved |
| Add latent consistency loss (ablation) | LE-PDE · 2D Navier-Stokes, ν=10^-5, T=20, N=1000 | relative L2 error: 0.2316 -> 0.1861 | Improved |
| Add reconstruction loss (ablation) | LE-PDE · 1D nonlinear PDE E2, (250,50) | accumulated error: 1.506 -> 1.127 | Improved |
| Add reconstruction loss (ablation) | LE-PDE · 2D Navier-Stokes, ν=10^-5, T=20, N=1000 | relative L2 error: 0.2025 -> 0.1861 | Improved |
| Use training horizon M=1 rather than M=4 (ablation) | LE-PDE · 1D nonlinear PDE E2, (250,50) | accumulated error: 1.127 -> 5.710 | Hurt |
| Use training horizon M=1 rather than M=4 (ablation) | LE-PDE · 2D Navier-Stokes, ν=10^-5, T=20, N=1000 | relative L2 error: 0.1861 -> 0.2860 | Hurt |
| Use training horizon M=3 rather than M=4 (ablation) | LE-PDE · 1D nonlinear PDE E2, (250,50) | accumulated error: 1.127 -> 1.234 | Hurt |
Closest papers by meaning
- Uncertainty Quantification for Forward and Inverse Problems of PDEs via Latent Global Evolution (AAAI 2024 · similarity 0.91)
- Active Learning with Selective Time-Step Acquisition for PDEs (ICML 2025 · similarity 0.90)
- CALM-PDE: Continuous and Adaptive Convolutions for Latent Space Modeling of Time-dependent PDEs (NeurIPS 2025 · similarity 0.90)
- MultiPDENet: PDE-embedded Learning with Multi-time-stepping for Accelerated Flow Simulation (ICML 2025 · similarity 0.90)
- Stable Long-Horizon PDE Forecasting via Latent Structured Spectral Propagators (NeurIPS 2026 · similarity 0.89)
- Learned Lagrangian Models of PDEs via Euler–Lagrange Residual Minimization (NeurIPS 2026 · similarity 0.89)
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.
