JointFM: A Novel Approach to Stochastic Modeling

Despite advancements in Artificial Intelligence (AI), Stochastic Differential Equations (SDEs) remain a gold-standard formalism for modeling systems under uncertainty. However, applying SDEs practically presents significant challenges: risk modeling is high, calibration is complex, and high-fidelity simulations are computationally expensive.

JointFM inverts this paradigm. Instead of fitting SDEs to data, it samples an infinite stream of synthetic SDEs to train a generic model to directly predict future joint probability distributions. This approach establishes JointFM as the first foundation model for distributional predictions of coupled time series, requiring no task-specific calibration or fine-tuning.

Operating in a purely zero-shot setting, JointFM reduces energy loss by 14.2% relative to the strongest baseline when recovering oracle joint distributions generated by unseen synthetic SDEs.