PathFM uses flow matching for multivariate trajectory forecasting
October 14, 2026
PathFM employs flow matching to generate complete future paths, capturing the dependence structures missed by marginal quantile models. The model is distilled to one or two steps, allowing it to generate multiple trajectories in roughly the time of a single quantile query.
HOW THIS AFFECTS YOU
●
builderYou can use this for high-speed, multi-path forecasting of complex multivariate time series.
●
researcherThis demonstrates the utility of distilled flow matching for efficient generative trajectory modeling.