Robust Classification by Coupling Data Mollification with Label Smoothing
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912357289033728 |
|---|---|
| author | Heinonen, Markus Tran, Ba-Hien Kampffmeyer, Michael Filippone, Maurizio |
| author_facet | Heinonen, Markus Tran, Ba-Hien Kampffmeyer, Michael Filippone, Maurizio |
| contents | Introducing training-time augmentations is a key technique to enhance generalization and prepare deep neural networks against test-time corruptions. Inspired by the success of generative diffusion models, we propose a novel approach of coupling data mollification, in the form of image noising and blurring, with label smoothing to align predicted label confidences with image degradation. The method is simple to implement, introduces negligible overheads, and can be combined with existing augmentations. We demonstrate improved robustness and uncertainty quantification on the corrupted image benchmarks of CIFAR, TinyImageNet and ImageNet datasets. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_01494 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Robust Classification by Coupling Data Mollification with Label Smoothing Heinonen, Markus Tran, Ba-Hien Kampffmeyer, Michael Filippone, Maurizio Computer Vision and Pattern Recognition Machine Learning Introducing training-time augmentations is a key technique to enhance generalization and prepare deep neural networks against test-time corruptions. Inspired by the success of generative diffusion models, we propose a novel approach of coupling data mollification, in the form of image noising and blurring, with label smoothing to align predicted label confidences with image degradation. The method is simple to implement, introduces negligible overheads, and can be combined with existing augmentations. We demonstrate improved robustness and uncertainty quantification on the corrupted image benchmarks of CIFAR, TinyImageNet and ImageNet datasets. |
| title | Robust Classification by Coupling Data Mollification with Label Smoothing |
| topic | Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2406.01494 |