PathFM: Fast Generative Forecasting of Multivariate Trajectories
PathFM uses flow matching to forecast full multivariate trajectories and beats baselines on a new path-score benchmark.
PathFM is a generative time-series model that uses flow matching to produce complete multivariate future trajectories rather than only marginal quantiles. Distillation to one or two steps and a shared history-attention cache lets it draw multiple paths about as fast as one quantile-model query and supports direct calibration. It reports competitive marginal accuracy on fev-bench and TIME and a higher aggregate path score than evaluated baselines on the new Forecast Path Score Bench. On chaotic systems, the flow model sustains long-run variability while strong baselines collapse.
- PathFM uses flow matching to sample complete future trajectories, preserving dependence.
- One- or two-step distillation and a shared history cache speed multi-path generation.
- It is competitive on fev-bench and TIME and leads fps-bench path scores.
- On chaotic systems it maintains long-run variability while strong baselines collapse.
From the original article
1 min read at huggingface.co
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.
By Jiachen Yao, Xiyuan Zhang, Annan Yu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.