Mod\`eles de Fondation et Ajustement : Vers une Nouvelle G\'en\'eration de Mod\`eles pour la Pr\'evision des S\'eries Temporelles

TL;DR

This paper reviews foundation models for zero-shot time series forecasting, highlighting how pretraining and fine-tuning improve predictions on unseen datasets, especially for long-term forecasts.

Contribution

It introduces a comprehensive review of architectures and training strategies for foundation models in time series forecasting, emphasizing the benefits of fine-tuning.

Findings

Fine-tuning enhances zero-shot forecasting performance.

Large-scale models learn generalizable representations.

Performance improves notably for long-term horizons.

Abstract

Inspired by recent advances in large language models, foundation models have been developed for zero-shot time series forecasting, enabling prediction on datasets unseen during pretraining. These large-scale models, trained on vast collections of time series, learn generalizable representations for both point and probabilistic forecasting, reducing the need for task-specific architectures and manual tuning. In this work, we review the main architectures, pretraining strategies, and optimization methods used in such models, and study the effect of fine-tuning after pretraining to enhance their performance on specific datasets. Our empirical results show that fine-tuning generally improves zero-shot forecasting capabilities, especially for long-term horizons.