Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Infinite-Fidelity Coregionalization for Physical Simulation
Shibo Li, Zheng Wang, Robert Kirby, Shandian Zhe
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 |
|---|---|---|---|
| model continuous fidelity with IFC-ODE^2 versus restricting to the highest training fidelity m=1 (extrapolate to 128x128 mesh) | IFC-ODE^2 · Poisson | nRMSE: 0.036 -> 0.018 | Improved |
| extrapolate IFC-ODE^2 beyond training fidelity to m=2.14 | IFC-ODE^2 · Heat | nRMSE: 0.074 -> 0.061 | Improved |
| use GP prior over bases (IFC-GPODE) instead of latent ODE (IFC-ODE^2) for fidelities beyond training range m>1 (ablation) | IFC · Poisson and Heat, extrapolation m>1 | nRMSE | Hurt |
| use GP prior (IFC-GPODE) instead of ODE^2 within training fidelity range, 0<=m<=1 (ablation) | IFC · Poisson and Heat, interpolation | nRMSE | Improved |
| replace single-fidelity SF degeneration with infinite-fidelity IFC (ablation) | IFC · Poisson, Heat, Burgers | nRMSE | Improved |
| IFC versus DRC, MFHoGP, DMF discrete-fidelity baselines | IFC-GPODE and IFC-ODE^2 · Poisson, Heat, Burgers, topology optimization, fluid dynamics | nRMSE | Improved |
| increase latent output dimension K from 5 to 20 (ablation) | IFC · Topology optimization | nRMSE | Improved |
| use more hidden layers than two in NN components (ablation) | IFC and DMF · PDE benchmarks | nRMSE | No clear effect |
| replace tanh with ReLU or LeakyReLU activation (ablation) | IFC · PDE benchmarks | nRMSE | Hurt |
| use ODE-based IFC instead of discrete-fidelity DMF/DRC for training cost | IFC-ODE^2 · Poisson training time, K=20 | seconds per epoch: 0.04 -> 7.84 | Hurt |
| use ODE-based IFC instead of DMF for training cost | IFC-ODE^2 · Fluid dynamics training time, K=20 | seconds per epoch: 0.10 -> 21.54 | Hurt |
Closest papers by meaning
- Compute-optimal data scaling for neural surrogates via multi-fidelity training (NeurIPS 2026 · similarity 0.87)
- Arbitrarily-Conditioned Multi-Functional Diffusion for Multi-Physics Emulation (ICML 2025 · similarity 0.86)
- Calibrated Physics-Informed Uncertainty Quantification (ICML 2025 · similarity 0.85)
- Space and time continuous physics simulation from partial observations (ICLR 2024 · similarity 0.85)
- Neural Physical Simulation with Multi-Resolution Hash Grid Encoding (AAAI 2024 · similarity 0.84)
- Breaking the Discretization Barrier of Continuous Physics Simulation Learning (NeurIPS 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.
