Time-series forecasting through the lens of dynamics

arXiv:2507.15774v2 Announce Type: replace 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 exttt{PRO-DYN} nomenclature to analyze existing models through the lens of dynamics. Two observations thus emerge: extbf{1.} under-performing architectures learn dynamics at most partially, extbf{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.

Source: arXiv cs.LG | 2026-05-05