[论文] PathFM: Fast Generative Forecasting of Multivariate Trajectories

论文概要

研究领域: ML 作者: Jiachen Yao, Xiyuan Zhang, Annan Yu, Boran Han, Abdul Fatir Ansari, Shuai Zhang, Michael Bohlke-Schneider, Bernie Wang 发布时间: 2026-10-15 arXiv: 2610.22358

中文摘要

时间序列基础模型在通用预测中越来越受欢迎，但大多数领先模型预测边缘分位数，未来轨迹的依赖结构仍未指定。这种区别对累计需求、峰值负载和事件时间等轨迹依赖量很重要。为了建模这种依赖，我们引入 PathFM，这是一个基于轨迹的预测模型，使用流匹配生成完整的未来路径。我们进一步将模型蒸馏到一到两步，并在采样间共享历史注意力缓存，使 PathFM 能在大约相当于分位数模型一次查询的时间内生成多条路径。PathFM 在预测基准 fev-bench 和 TIME 上实现了具有竞争力的边缘精度。我们引入预测路径评分基准（fps-bench），PathFM 在其上取得比所有评估基线更高的综合路径分数。混沌系统实验表明，我们的流模型维持长期变异性，而强基线迅速崩溃。PathFM 使生成式时间序列预测既具有竞争力又实用。

原文摘要

Time series foundation models are becoming increasingly popular for general-purpose forecasting. However, most leading models predict marginal quantiles, leaving the dependence structure of future trajectories unspecified. This distinction matters for trajectory-dependent quantities such as cumulative demand, peak load, and event timing, yet point and marginal scores do not reveal it. To model this dependence, we introduce PathFM, a trajectory-based forecasting model that uses flow matching to generate complete future paths. We further distill the model to one or two steps and share history attention cache across draws. Together, these enable PathFM to generate multiple paths for a single series in roughly the time needed for one query to a quantile model. The distilled sampler also makes ...

#论文 #arXiv #ML #小凯