TempoPFN: Synthetic Pre-training of Linear RNNs for Zero-shot Time Series Forecasting

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Moroshan, Vladyslav, Siems, Julien, Zela, Arber, Carstensen, Timur, Hutter, Frank
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908815203500032
author Moroshan, Vladyslav
Siems, Julien
Zela, Arber
Carstensen, Timur
Hutter, Frank
author_facet Moroshan, Vladyslav
Siems, Julien
Zela, Arber
Carstensen, Timur
Hutter, Frank
contents Foundation models for zero-shot time series forecasting face challenges in efficient long-horizon prediction and reproducibility, with existing synthetic-only approaches underperforming on challenging benchmarks. This paper presents TempoPFN, a univariate time series foundation model based on linear Recurrent Neural Networks (RNNs) pre-trained exclusively on synthetic data. The model uses a GatedDeltaProduct architecture with state-weaving for fully parallelizable training across sequence lengths, eliminating the need for windowing or summarization techniques while maintaining robust temporal state-tracking. Our comprehensive synthetic data pipeline unifies diverse generators, including stochastic differential equations, Gaussian processes, and audio synthesis, with novel augmentations. In zero-shot evaluations on the Gift-Eval, fev-bench and Chronos-ZS benchmarks, TempoPFN achieves top-tier competitive performance, outperforming all existing synthetic-only approaches and surpassing the majority of models trained on real-world data, while being more efficient than existing baselines by leveraging fully parallelizable training and inference. We open-source our complete data generation pipeline and training code, providing a reproducible foundation for future research.
format Preprint
id arxiv_https___arxiv_org_abs_2510_25502
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle TempoPFN: Synthetic Pre-training of Linear RNNs for Zero-shot Time Series Forecasting
Moroshan, Vladyslav
Siems, Julien
Zela, Arber
Carstensen, Timur
Hutter, Frank
Machine Learning
Artificial Intelligence
Foundation models for zero-shot time series forecasting face challenges in efficient long-horizon prediction and reproducibility, with existing synthetic-only approaches underperforming on challenging benchmarks. This paper presents TempoPFN, a univariate time series foundation model based on linear Recurrent Neural Networks (RNNs) pre-trained exclusively on synthetic data. The model uses a GatedDeltaProduct architecture with state-weaving for fully parallelizable training across sequence lengths, eliminating the need for windowing or summarization techniques while maintaining robust temporal state-tracking. Our comprehensive synthetic data pipeline unifies diverse generators, including stochastic differential equations, Gaussian processes, and audio synthesis, with novel augmentations. In zero-shot evaluations on the Gift-Eval, fev-bench and Chronos-ZS benchmarks, TempoPFN achieves top-tier competitive performance, outperforming all existing synthetic-only approaches and surpassing the majority of models trained on real-world data, while being more efficient than existing baselines by leveraging fully parallelizable training and inference. We open-source our complete data generation pipeline and training code, providing a reproducible foundation for future research.
title TempoPFN: Synthetic Pre-training of Linear RNNs for Zero-shot Time Series Forecasting
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2510.25502