Learning Less Generalizable Patterns for Better Test-Time Adaptation
Résumé
Deep neural networks often fail to generalize outside of their training distribution, particularly when only a single data domain is available during training. While test-time adaptation has yielded encouraging results in this setting, we argue that to reach further improvements, these approaches should be combined with training procedure modifications aiming to learn a more diverse set of patterns. Indeed, test-time adaptation methods usually have to rely on a limited representation because of the shortcut learning phenomenon: only a subset of the available predictive patterns is learned with standard training. In this paper, we first show that the combined use of existing training-time strategies and test-time batch normalization, a simple adaptation method, does not always improve upon the test-time adaptation alone on the PACS benchmark. Furthermore, experiments on Office-Home show that very few training-time methods improve upon standard training, with or without test-time batch normalization. Therefore, we propose a novel approach that mitigates the shortcut learning behavior by having an additional classification branch learn less predictive and generalizable patterns. Our experiments show that our method improves upon the state-of-the-art results on both benchmarks and benefits the most to test-time batch normalization.
Fichier principal
egpaper.pdf (313.85 Ko)
Télécharger le fichier
L2GP_VISIGRAPP2023_arXiv_version.zip (235.73 Ko)
Télécharger le fichier
egpaper.bbl (10.97 Ko)
Télécharger le fichier
egpaper.log (23.54 Ko)
Télécharger le fichier
egrebuttal.bbl (172 B)
Télécharger le fichier
egrebuttal.log (16.72 Ko)
Télécharger le fichier
egrebuttal.pdf (110.38 Ko)
Télécharger le fichier
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|
