Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Physics-Informed Residual Flows
Jephte Abijuru, Mayank Nagda, Phil Ostheimer, Sebastian Josef Vollmer, Marius Kloft, Sophie Fellenz
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 attention in PINNsFormer with a linear projection (ablation) | PINNsFormer · Convection PDE | rMAE: 0.510 -> 0.012 | Improved |
| Replace attention in PINNsFormer with a shallow MLP (ablation) | PINNsFormer · Wave PDE | rMAE: 0.270 -> 0.142 | Improved |
| Remove the SSM from PINNMamba (ablation) | PINNMamba · Convection PDE | rMAE: 0.019 -> 0.012 | Improved |
| Replace SSM with parameter-matched MLP in PINNMamba (ablation) | PINNMamba · Convection PDE | rMAE: 0.019 -> 0.063 | Hurt |
| Remove SSM from PINNMamba (ablation) | PINNMamba · Reaction and Wave PDE | rMAE: 0.010 -> 0.013 (Reaction); 0.020 -> 0.029 (Wave) | Hurt |
| Stack residual updates throughout the network (ResPINN) instead of one or two residual blocks | PINN · Convection PDE | rMAE: 0.8514 -> 0.0028 | Improved |
| Use ResPINN instead of PirateNet | Heat PDE · Heat PDE | rMAE: 0.0005 -> 0.0035 | Hurt |
| Use continuous-depth (O-PINN) instead of discrete ResPINN with wavelet activation (ablation) | ResPINN · Reaction PDE | rMAE: 0.008 -> 0.003 | Improved |
| Use wavelet instead of tanh activation in ResPINN (ablation) | ResPINN · Wave PDE | rMAE: 0.030 -> 0.070 | Hurt |
| Use wavelet instead of tanh activation in ResPINN (ablation) | ResPINN · Convection PDE | rMAE: 0.015 -> 0.006 | Improved |
| Add residual connections after every two layers (same parameter count) (ablation) | PINN · 1D convection Jacobian spectrum | singular values | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PINNsFormer — PINNsFormer: A Transformer-Based Framework For Physics-Informed Neural Networks (ICLR 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Per-Loss Adapters for Gradient Conflict in Physics-Informed Neural Networks (NeurIPS 2026 · similarity 0.89)
- Gradient Alignment in Physics-informed Neural Networks: A Second-Order Optimization Perspective (NeurIPS 2025 · similarity 0.89)
- Efficient Error Certification for Physics-Informed Neural Networks (ICML 2024 · similarity 0.89)
- ConFIG: Towards Conflict-free Training of Physics Informed Neural Networks (ICLR 2025 · similarity 0.89)
- Sub-Sequential Physics-Informed Learning with State Space Model (ICML 2025 · similarity 0.88)
- RoPINN: Region Optimized Physics-Informed Neural Networks (NeurIPS 2024 · similarity 0.88)
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.
