← sources

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

source: reddit (reddit:MachineLearning)published 2026-10-18 16:43ingested 2026-10-19 02:39

retained text

*Accepted at NeurIPS 2026*

TLDR: The paper presents a library of synthetic time series generators and augmentations to train time series foundation models, in their case a DeltaProduct based linear RNN. The synthetic data has already been used in Toto-2 and TS-ICL.

*Arxiv:* [https://arxiv.org/abs/2510.25502](https://arxiv.org/abs/2510.25502)  
*GitHub:* [https://github.com/automl/TempoPFN](https://github.com/automl/TempoPFN)  
*Huggingface*: [https://huggingface.co/AutoML-org/TempoPF](https://huggingface.co/AutoML-org/TempoPFN)[N](https://huggingface.co/AutoML-org/TempoPFN)

Authors: Vladyslav Moroshan, Julien Siems, Arber Zela, Timur Carstensen, Frank Hutter

document fetched 2026-10-19 00:25 · canonical https://reddit.com/r/MachineLearning/comments/1x97v19/tempopfn_synthetic_pretraining_of_linear_rnns_for