Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Unveiling Multi-regime Patterns in SciML: Distinct Failure Modes and Regime-specific Optimization
Yuxin Wang, Yuanzhe Hu, Xiaokun Zhong, Xiaopeng Wang, Haiquan Lu, Tianyu Pang, Michael W. Mahoney, Yujun Yan, Pu Ren, Yaoqing Yang
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 hard-constrained ALM instead of soft penalty (ablation) | PINN · 1D convection | trainability boundary beta: 25 -> 50 | Improved |
| use hard-constrained ALM (ablation) | PINN · 1D convection Regime III | generalization gap | No clear effect |
| add curriculum learning (ablation) | physics-constrained SciML (PINN, PINO, PINODE) · regime maps | Regime I/II boundary | Improved |
| use curriculum learning instead of ALM (ablation) | PINN · 1D convection beta=30 | collocation points needed: 5000 -> 1000 | Improved |
| add RoPINN stabilization (ablation) | PINN · 1D convection | trainability boundary | Improved |
| add RoPINN stabilization (ablation) | PINN · 1D convection Regime III | generalization failure | No clear effect |
| add NNCG fine-tuning after L-BFGS (ablation) | PINN · 1D convection Regime I | test error: about 50% reduction | Improved |
| add NNCG fine-tuning after L-BFGS (ablation) | PINN · 1D convection Regime I | training loss: roughly 60% improvement | Improved |
| add NNCG fine-tuning after L-BFGS (ablation) | PINN · high-beta Regime II/III | training and test error | No clear effect |
Closest papers by meaning
- On the Surprising Effectiveness of Large Learning Rates under Standard Width Scaling (NeurIPS 2025 · similarity 0.85)
- Three Mechanisms of Feature Learning in a Linear Network (ICLR 2025 · similarity 0.85)
- The Butterfly Effect: Neural Network Training Trajectories Are Highly Sensitive to Initial Conditions (ICML 2025 · similarity 0.85)
- Neural Collapse Beyond the Unconstrained Features Model: Landscape, Dynamics, and Generalization in the Mean-Field Regime (ICML 2025 · similarity 0.84)
- Scaling Law with Learning Rate Annealing (NeurIPS 2025 · similarity 0.84)
- Scaling Collapse Reveals Universal Dynamics in Compute-Optimally Trained Neural Networks (ICML 2025 · similarity 0.84)
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.
