Research map / Probabilistic and scientific modeling / Dynamical systems
Physics-Informed Functional Tucker Method with RKHS Factors for Sparse Spatiotemporal Reconstruction
Zhengyang Liu, Jianguo Huang, Junxi Hu, Yucheng Mi, Haojie Cai, Wenxuan Shen, zhikun zhang, Renfeng Peng, Guangtao Zhang, Zhiqiang Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a physics-informed functional Tucker method with RKHS factors for reconstructing physical fields from sparse, irregular spatiotemporal observations.
Problem
Reconstructing physical dynamics from sparse and irregular spatiotemporal observations is ill-posed: models can fit sensors while producing nonphysical off-sensor structures or unreliable predictions between support times.
Approach
Combines a functional Tucker decomposition with reproducing kernel Hilbert space (RKHS) factor functions and physics-informed constraints to regularize the reconstruction away from sensors (based on the title and abstract; details not available).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add PDE residual loss at collocation points (ablation) | PhysFTM · NS2D | off-sensor RMSE: 0.086781 -> 0.061196 | Improved |
| add PDE residual loss (ablation) | PhysFTM · AC2D | off-sensor RMSE: 0.092520 -> 0.089395 | Improved |
| add RKHS norm regularization (ablation) | PhysFTM · NS2D | off-sensor RMSE: 0.115421 -> 0.061196 | Improved |
| alternate core solve with gradient factor learning vs solve-only (ablation) | PhysFTM · AC2D | off-sensor RMSE: 0.173653 -> 0.089395 | Improved |
| alternate core solve with gradient factor learning vs gradient-only (ablation) | PhysFTM · AC2D | off-sensor RMSE: 0.352132 -> 0.089395 | Improved |
| replace MLP mode factors with RKHS factors (ablation) | FTM · AC2D | off-sensor RMSE support frames: 0.1026 -> 0.0925 | Improved |
| impose PDE residual at off-sensor collocation points rather than sensor or full-domain points (ablation) | PhysFTM · AC2D | RMSE: 0.0898 -> 0.0884 | Improved |
| add learned ODE-RNN correction to cubic Hermite interpolation of cores (ablation) | PhysFTM · NS2D (s=0.03) | non-grid-aligned RMSE: 0.063790 -> 0.055546 | Improved |
| PhysFTM instead of RecFNO on sensor locations in missing frames | PhysFTM · AC2D | sensor RMSE missing frames: 0.02178 vs 0.05035 | Hurt |
| PhysFTM instead of LRTFR on Ocean3D interval-missing | PhysFTM · Ocean3D | off-sensor RMSE missing frames: 0.18377 vs 0.18565 | Hurt |
| PhysFTM vs other methods on divergence | PhysFTM · NS2D | divergence error: 0.134170 vs PINNs 0.120243 | Mixed |
Closest papers by meaning
- Generating Full-field Evolution of Physical Dynamics from Irregular Sparse Observations (NeurIPS 2025 · similarity 0.89)
- StreamPhy: Streaming Inference of High-Dimensional Physical Dynamics via State Space Models (NeurIPS 2026 · similarity 0.88)
- Sparsely-Supervised Data Assimilation via Physics-Informed Schrödinger Bridge (NeurIPS 2026 · similarity 0.86)
- DiffATS: Diffusion in Aligned Tensor Space (NeurIPS 2026 · similarity 0.85)
- Functional Bayesian Tucker Decomposition for Continuous-indexed Tensor Data (ICLR 2024 · similarity 0.84)
- Continuous Field Reconstruction from Sparse Observations with Implicit Neural Networks (ICLR 2024 · 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.
