Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
TORNADO: Adaptive Latent Stochastic Transport for Calibrated Probabilistic PDE Forecasting
Jorge Mifsut Benet, Armand Kassaï Koupaï, Ramon Daniel Regueiro-espino, Nicolas Baskiotis, Patrick Gallinari
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 TORNADO, a latent probabilistic forecasting framework for PDE surrogates that uses adaptive, state-dependent stochastic transport to produce calibrated ensemble forecasts in few-step inference.
Problem
Probabilistic neural PDE surrogates often are miscalibrated for uncertainty quantification, especially under compute-constrained, few-step inference; existing generative methods use fixed global stochasticity that cannot adapt to local dynamical state.
Approach
Replaces fixed global stochasticity with adaptive latent stochastic transport that adjusts predictive uncertainty to the local dynamical state, aiming for calibration under few-step sampling.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use learned state-dependent diffusion instead of fixed schedule (ablation) | TORNADO sampler · Kolmogorov flow | RMSCE: 0.067 -> 0.039 | Improved |
| Use learned state-dependent diffusion instead of constant amplitude (ablation) | TORNADO sampler · Kolmogorov flow | RMSCE: 0.085 -> 0.039 | Improved |
| Add stochastic diffusion to deterministic Euler sampling (ablation) | TORNADO sampler · Kolmogorov flow | |log SSR|: 9.674 -> 0.264 | Improved |
| Use constant diffusion amplitude instead of learned diffusion (ablation) | TORNADO sampler · Kolmogorov flow | MSE: 0.371 -> 2.701 | Hurt |
| Use TORNADO instead of stochastic interpolants | latent PDE forecaster · Kolmogorov flow | RMSCE (x1e-2): 10.56 -> 3.95 | Improved |
| Use TORNADO instead of stochastic interpolants | latent PDE forecaster · KS | |log SSR|: 0.1417 -> 0.0209 | Improved |
| Use TORNADO instead of flow matching | latent PDE forecaster · Kolmogorov flow | MSE (x1e-3): 107 -> 120 | Hurt |
| Use TORNADO instead of flow matching | latent PDE forecaster · Kolmogorov flow | RMSCE (x1e-2): 20.33 -> 3.95 | Improved |
| Use TORNADO instead of stochastic interpolants | latent PDE forecaster · ERA5 | RMSCE (x1e-2): 6.39 -> 5.77 | Improved |
| Use TORNADO instead of stochastic interpolants | latent PDE forecaster · Burgers | MSE (x1e-3): 0.676 -> 0.764 | Hurt |
Closest papers by meaning
- Probabilistic Forecasting with Stochastic Interpolants and Föllmer Processes (ICML 2024 · similarity 0.87)
- Stable Long-Horizon PDE Forecasting via Latent Structured Spectral Propagators (NeurIPS 2026 · similarity 0.87)
- Neural Stochastic Flows: Solver-Free Modelling and Inference for SDE Solutions (NeurIPS 2025 · similarity 0.87)
- First-Order Trajectory Matching: Fast Ensemble Predictions of Chaotic, Turbulent, Stochastic Systems (NeurIPS 2026 · similarity 0.86)
- Elucidated Rolling Diffusion Models for Probabilistic Forecasting of Complex Dynamics (NeurIPS 2025 · similarity 0.85)
- SPDEBench: An Extensive Benchmark for Learning Stochastic PDEs (NeurIPS 2026 · similarity 0.85)
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.
