Research map / Probabilistic and scientific modeling / Differential equations and PDEs
PirateNets: Physics-informed Deep Learning with Residual Adaptive Networks
Sifan Wang, Bowen Li, Yuhan Chen, Paris Perdikaris
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 |
|---|---|---|---|
| Initialize adaptive residual parameter alpha to 0 instead of 1 (ablation) | PirateNet · Grey-Scott | relative L2 error: one or two orders of magnitude lower error | Improved |
| Initialize adaptive residual parameter alpha to 0 instead of 1 (ablation) | PirateNet · Ginzburg-Landau | relative L2 error: alpha=1 gives complete model collapse for deeper networks | Improved |
| Add gating operations to PirateNet blocks (ablation) | PirateNet · Grey-Scott | relative L2 error | Improved |
| Add gating operations to PirateNet blocks (ablation) | PirateNet · Ginzburg-Landau | relative L2 error | Improved |
| Increase depth of PirateNet (ablation) | PirateNet · Allen-Cahn | relative L2 error: consistent improvement in accuracy as depth increases | Improved |
| Increase depth of Modified MLP (ablation) | Modified MLP · Allen-Cahn | relative L2 error: increasing error in deeper networks | Hurt |
| Increase depth of MLP (ablation) | MLP PINN · Allen-Cahn | relative L2 error: almost 100% error at 18 hidden layers | Hurt |
| Add residual connections to MLP (ablation) | MLP PINN · Allen-Cahn | relative L2 error | No clear effect |
| Add physics-informed least-squares initialization of the last layer (ablation) | Modified MLP and PirateNet · Allen-Cahn | relative L2 error | Improved |
| Fit linearized PDE solution instead of initial condition for initialization (ablation) | PirateNet · Allen-Cahn | relative L2 error | No clear effect |
| Use PirateNet instead of Modified MLP backbone without physics-informed initialization (ablation) | PirateNet · Korteweg-De Vries | relative L2 error | Improved |
| Increase PirateNet depth beyond 20 layers (ablation) | PirateNet · Lid-driven cavity flow (Re=3200) | relative L2 error: saturates beyond depth 20, similar to depth 18 | No clear effect |
Closest papers by meaning
- Efficient Error Certification for Physics-Informed Neural Networks (ICML 2024 · similarity 0.88)
- RoPINN: Region Optimized Physics-Informed Neural Networks (NeurIPS 2024 · similarity 0.88)
- Sub-Sequential Physics-Informed Learning with State Space Model (ICML 2025 · similarity 0.87)
- Gradient Alignment in Physics-informed Neural Networks: A Second-Order Optimization Perspective (NeurIPS 2025 · similarity 0.87)
- Challenges in Training PINNs: A Loss Landscape Perspective (ICML 2024 · similarity 0.87)
- PINN Balls: Scaling Second-Order Methods for PINNs with Domain Decomposition and Adaptive Sampling (NeurIPS 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.
