Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Decoupling Universal Laws and Environmental Heterogeneity: A Physics-Inspired Framework for Robust Spatio-Temporal Forecasting
Aoyu Liu, Liming Wei, Yaying Zhang
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 the PDE solver with an MLP (ablation) | STPDE · CA-D3 (ID and OOD) | MAE | Hurt |
| add the Invariant Diffusion Operator (ablation) | STPDE · CA-D3 and LA (ID and OOD) | MAE | Improved |
| add the Environment Basis Manifold (ablation) | STPDE · CA-D3 and LA (ID and OOD) | MAE | Improved |
| add stochastic environment perturbation (ablation) | STPDE · CA-D3 and LA (OOD) | MAE | Improved |
| use a GCN instead of the linear-attention Invariant Diffusion Operator (ablation) | STPDE · CA-D3 and LA | MAE | Hurt |
| increase hidden dimension D from 64 to 256, then 320 (ablation) | STPDE · CA-D3, LA | MAE | Mixed |
| raise perturbation ratio to about 25% or lower to 5% (ablation) | STPDE · CA-D3, LA | MAE | Hurt |
| increase number of environment bases K beyond 4 (ablation) | STPDE · CA-D3, LA | MAE | No clear effect |
| use STPDE in place of strongest baseline for in-distribution forecasting | ID forecasting · LargeST four regions | MAE (averaged): 1.5% reduction | Improved |
| use STPDE in place of strongest baseline for OOD generalization | OOD forecasting · LargeST four regions | MAE (averaged): 4.0% reduction | Improved |
| adapt by reconstructing only the Environment Basis Manifold on 10% target data | STPDE · CA-D3 few-shot transfer | MAE: CrossST 17.51 -> 16.25 | Improved |
| use OOD-specific methods (STONE, STOP) instead of ID-oriented GWNet | OOD generalization baselines · LargeST | MAE: CA-D3 OOD: GWNet 16.84, STONE 17.89, STOP 17.68 | Hurt |
Closest papers by meaning
- Causal Deciphering and Inpainting in Spatio-Temporal Dynamics via Diffusion Model (NeurIPS 2024 · similarity 0.88)
- Elucidated Rolling Diffusion Models for Probabilistic Forecasting of Complex Dynamics (NeurIPS 2025 · similarity 0.87)
- Stable Long-Horizon PDE Forecasting via Latent Structured Spectral Propagators (NeurIPS 2026 · similarity 0.87)
- Predicting partially observable dynamical systems via diffusion models with a multiscale inference scheme (NeurIPS 2025 · similarity 0.87)
- Learning with Calibration: Exploring Test-Time Computing of Spatio-Temporal Forecasting (NeurIPS 2025 · similarity 0.86)
- ABC: Any-Subset Autoregression via Non-Markovian Diffusion Bridges in Continuous Time and Space (NeurIPS 2026 · similarity 0.86)
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.
