Research map / Probabilistic and scientific modeling / Dynamical systems
Neural Implicit Flow: a mesh-agnostic dimensionality reduction paradigm of spatio-temporal data
Shaowu Pan, Steven L. Brunton, J. Nathan Kutz
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 NIF (ShapeNet/ParameterNet) instead of a standard MLP, Swish activation | parametric surrogate · 1D parametric K-S | Test RMSE: 1.10 -> 0.64 | Improved |
| use NIF instead of MLP with tanh activation | parametric surrogate · 1D parametric K-S | Test RMSE: 1.17 -> 0.99 | Improved |
| replace tanh with Swish activation in NIF (ablation) | NIF · 1D parametric K-S | Test RMSE: 0.99 -> 0.64 | Improved |
| replace tanh with Swish activation in MLP (ablation) | MLP · 1D parametric K-S | Test RMSE: 1.17 -> 1.10 | Improved |
| use NIF instead of MLP when training data size varies (15, 24, 29 parameter samples) (ablation) | NIF-Swish · 1D parametric K-S | Test RMSE: approximately half of the training data | Improved |
| use NIF instead of MLP at equal parameter count (7000 to 34000) (ablation) | NIF-Swish · 1D parametric K-S | Test RMSE: lowers the testing error by half | Improved |
| use NIF instead of CAE on adaptive mesh data | NIF autoencoder · 2D Rayleigh-Taylor | RMSE: CAE 10, 5, 1.7 times more error than NIF | Improved |
| use NIF instead of SVD for rank 1 reduction | NIF autoencoder · 2D Rayleigh-Taylor | RMSE | Improved |
| use NIF instead of SIREN at small width | NIF · 2D turbulence slice x-velocity | normalized error: 0.121 -> 0.071 (width 36); 0.040 -> 0.022 (width 75) | Improved |
| use NIF instead of SIREN at width 150 | NIF · 2D turbulence slice x-velocity | normalized error: 0.0116 vs 0.0131 | No clear effect |
| use NIF instead of Fourier feature networks | NIF · 2D turbulence slice x-velocity | normalized error: 0.263 -> 0.071 (width 36) | Improved |
| separate spatial from time input (NIF vs SIREN taking t,x,y,z) | NIF · 3D HIT spatial query | network size / cost at same error | Improved |
Closest papers by meaning
- PEINR: A Physics-enhanced Implicit Neural Representation for High-Fidelity Flow Field Reconstruction (ICML 2025 · similarity 0.88)
- FlowMixer: A Depth-Agnostic Neural Architecture for Interpretable Spatiotemporal Forecasting (NeurIPS 2025 · similarity 0.86)
- Learning Spatiotemporal Dynamical Systems from Point Process Observations (ICLR 2025 · similarity 0.85)
- SplatFlow: Self-Supervised Dynamic Gaussian Splatting in Neural Motion Flow Field for Autonomous Driving (CVPR 2025 · similarity 0.85)
- Gridded Transformer Neural Processes for Spatio-Temporal Data (ICML 2025 · similarity 0.85)
- CALM-PDE: Continuous and Adaptive Convolutions for Latent Space Modeling of Time-dependent PDEs (NeurIPS 2025 · 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.
