Research
158d agoJointFM: Foundation Model for Joint Distribution Prediction
JointFM, a new foundation model, learns to predict joint probability distributions for coupled time series by sampling synthetic stochastic differential equations. This approach bypasses the brittle calibration of traditional SDEs, enabling scalable, high‑fidelity forecasts across finance, climate, and engineering.
The Signal
By training a generic model on an infinite stream of synthetic data, JointFM offers a globally applicable tool for uncertainty quantification.