Research map / Probabilistic and scientific modeling / Differential equations and PDEs
PINNfluence: Interpreting PINNs through Influence Functions
Aleksander Krasowski, Jonas Naujoks, Moritz Weckbecker, Galip Ümit Yolcu, Thomas Wiegand, Sebastian Lapuschkin, Wojciech Samek, René Pascal Klausen
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 the inverse-Hessian factor (influence function) instead of gradient dot product (ablation) | PINNfluence · Heat, PBRF validation, poorly-trained | Spearman correlation with PBRF: 0.09 -> 0.45 | Improved |
| use the inverse-Hessian factor instead of gradient dot product (ablation) | PINNfluence · Burgers', PBRF validation, well-trained | Spearman correlation with PBRF: -0.08 -> 0.77 | Improved |
| use the inverse-Hessian factor instead of gradient dot product (ablation) | PINNfluence · Allen-Cahn, PBRF validation, poorly-trained | Spearman correlation with PBRF: -0.01 -> 0.69 | Improved |
| use the inverse-Hessian factor instead of gradient dot product (ablation) | PINNfluence · Drift-Diffusion, PBRF validation, poorly-trained | Spearman correlation with PBRF: 0.07 -> 0.08 | Mixed |
| use influence estimates to predict exact leave-one-out retraining effects | PINNfluence · Heat, leave-one-out validation, well-trained | Spearman correlation with LOO: 0.00 | No clear effect |
| well-trained versus poorly-trained PINN, temporal influence indicator | PINNfluence · Wave, temporal influence | temporal influence indicator: 0.41 -> 0.11 | Mixed |
| well-trained versus poorly-trained PINN, temporal influence indicator | PINNfluence · Drift-Diffusion, temporal influence | temporal influence indicator: 0.46 -> 0.21 | Mixed |
Closest papers by meaning
- PINNsFormer: A Transformer-Based Framework For Physics-Informed Neural Networks (ICLR 2024 · similarity 0.90)
- Efficient Error Certification for Physics-Informed Neural Networks (ICML 2024 · similarity 0.89)
- When and Why Adversarial Training Improves PINNs: A Neural Tangent Kernel Perspective (NeurIPS 2026 · similarity 0.89)
- Challenges in Training PINNs: A Loss Landscape Perspective (ICML 2024 · similarity 0.88)
- Sub-Sequential Physics-Informed Learning with State Space Model (ICML 2025 · similarity 0.88)
- PINN Balls: Scaling Second-Order Methods for PINNs with Domain Decomposition and Adaptive Sampling (NeurIPS 2025 · 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.
