On the Role of Reversible Instance Normalization
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866908881887690752 |
|---|---|
| author | Berthelier, Gaspard Nabil, Tahar Naour, Etienne Le Niamke, Richard Perlaza, Samir Neglia, Giovanni |
| author_facet | Berthelier, Gaspard Nabil, Tahar Naour, Etienne Le Niamke, Richard Perlaza, Samir Neglia, Giovanni |
| contents | Data normalization is a crucial component of deep learning models, yet its role in time series forecasting remains insufficiently understood. In this paper, we identify three central challenges for normalization in time series forecasting: temporal input distribution shift, spatial input distribution shift, and conditional output distribution shift. In this context, we revisit the widely used Reversible Instance Normalization (RevIN), by showing through ablation studies that several of its components are redundant or even detrimental. Based on these observations, we draw new perspectives to improve RevIN's robustness and generalization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_11869 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | On the Role of Reversible Instance Normalization Berthelier, Gaspard Nabil, Tahar Naour, Etienne Le Niamke, Richard Perlaza, Samir Neglia, Giovanni Machine Learning Data normalization is a crucial component of deep learning models, yet its role in time series forecasting remains insufficiently understood. In this paper, we identify three central challenges for normalization in time series forecasting: temporal input distribution shift, spatial input distribution shift, and conditional output distribution shift. In this context, we revisit the widely used Reversible Instance Normalization (RevIN), by showing through ablation studies that several of its components are redundant or even detrimental. Based on these observations, we draw new perspectives to improve RevIN's robustness and generalization. |
| title | On the Role of Reversible Instance Normalization |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2603.11869 |