📄 中文摘要
随着人工智能的快速发展，随机微分方程（SDEs）依然是建模不确定性系统的金标准。然而，在实践中应用SDEs面临诸多挑战：风险建模高，校准往往脆弱，高保真模拟计算成本昂贵。该研究提出了JointFM，一个颠覆这一范式的基础模型。JointFM通过对无限流的合成SDEs进行采样，训练通用模型以直接预测未来的联合概率分布，而不是将SDEs拟合到数据上。这一方法使JointFM成为首个用于耦合时间序列的分布预测的基础模型，且无需特定任务的适配。
JointFM-0.1: 多目标联合分布预测的基础模型
📄 English Summary
JointFM-0.1: A Foundation Model for Multi-Target Joint Distributional Prediction
The rapid advancements in Artificial Intelligence (AI) have not diminished the status of Stochastic Differential Equations (SDEs) as the gold standard for modeling systems under uncertainty. However, practical applications of SDEs face significant challenges, including high modeling risk, brittle calibration, and expensive high-fidelity simulations. This research introduces JointFM, a foundation model that inverts the traditional paradigm. Instead of fitting SDEs to data, JointFM samples an infinite stream of synthetic SDEs to train a generic model capable of directly predicting future joint probability distributions. This innovative approach establishes JointFM as the first foundation model for distributional predictions of coupled time series, eliminating the need for task-specific adaptations.