Time Series Forecasting Through the Lens of Dynamics

Alexis-Raja Brachet, Pierre-Yves Richard, Celine Hudelot
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:9494-9530, 2026.

Abstract

While deep learning is facing an homogenization across modalities led by Transformers, they are still challenged by shallow linear models in the time series forecasting task. Our hypothesis is that models should learn a direct link from past to future data points, which we identify as a learning dynamics capability. We develop an original $\texttt{PRO-DYN}$ nomenclature to analyze existing models through the lens of dynamics. Two observations thus emerge: 1. under-performing architectures learn dynamics at most partially, 2. the location of the dynamics block at the model end is of prime importance. Our systemic and empirical studies both confirm our observations on a set of performance-varying models with diverse backbones. We propose a simple plug-and-play methodology guiding model designs and improvements.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-brachet26a, title = {Time Series Forecasting Through the Lens of Dynamics}, author = {Brachet, Alexis-Raja and Richard, Pierre-Yves and Hudelot, Celine}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {9494--9530}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/brachet26a/brachet26a.pdf}, url = {https://proceedings.mlr.press/v306/brachet26a.html}, abstract = {While deep learning is facing an homogenization across modalities led by Transformers, they are still challenged by shallow linear models in the time series forecasting task. Our hypothesis is that models should learn a direct link from past to future data points, which we identify as a learning dynamics capability. We develop an original $\texttt{PRO-DYN}$ nomenclature to analyze existing models through the lens of dynamics. Two observations thus emerge: 1. under-performing architectures learn dynamics at most partially, 2. the location of the dynamics block at the model end is of prime importance. Our systemic and empirical studies both confirm our observations on a set of performance-varying models with diverse backbones. We propose a simple plug-and-play methodology guiding model designs and improvements.} }
Endnote
%0 Conference Paper %T Time Series Forecasting Through the Lens of Dynamics %A Alexis-Raja Brachet %A Pierre-Yves Richard %A Celine Hudelot %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-brachet26a %I PMLR %P 9494--9530 %U https://proceedings.mlr.press/v306/brachet26a.html %V 306 %X While deep learning is facing an homogenization across modalities led by Transformers, they are still challenged by shallow linear models in the time series forecasting task. Our hypothesis is that models should learn a direct link from past to future data points, which we identify as a learning dynamics capability. We develop an original $\texttt{PRO-DYN}$ nomenclature to analyze existing models through the lens of dynamics. Two observations thus emerge: 1. under-performing architectures learn dynamics at most partially, 2. the location of the dynamics block at the model end is of prime importance. Our systemic and empirical studies both confirm our observations on a set of performance-varying models with diverse backbones. We propose a simple plug-and-play methodology guiding model designs and improvements.
APA
Brachet, A., Richard, P. & Hudelot, C.. (2026). Time Series Forecasting Through the Lens of Dynamics. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:9494-9530 Available from https://proceedings.mlr.press/v306/brachet26a.html.

Related Material