Robustmix: Improving Robustness by Regularizing the Frequency Bias of Deep Nets
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912077030883328 |
|---|---|
| author | Ngnawe, Jonas Njifon, Marianne Abemgnigni Heek, Jonathan Dauphin, Yann |
| author_facet | Ngnawe, Jonas Njifon, Marianne Abemgnigni Heek, Jonathan Dauphin, Yann |
| contents | Deep networks have achieved impressive results on a range of well-curated benchmark datasets. Surprisingly, their performance remains sensitive to perturbations that have little effect on human performance. In this work, we propose a novel extension of Mixup called Robustmix that regularizes networks to classify based on lower-frequency spatial features. We show that this type of regularization improves robustness on a range of benchmarks such as Imagenet-C and Stylized Imagenet. It adds little computational overhead and, furthermore, does not require a priori knowledge of a large set of image transformations. We find that this approach further complements recent advances in model architecture and data augmentation, attaining a state-of-the-art mCE of 44.8 with an EfficientNet-B8 model and RandAugment, which is a reduction of 16 mCE compared to the baseline. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2304_02847 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Robustmix: Improving Robustness by Regularizing the Frequency Bias of Deep Nets Ngnawe, Jonas Njifon, Marianne Abemgnigni Heek, Jonathan Dauphin, Yann Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning Deep networks have achieved impressive results on a range of well-curated benchmark datasets. Surprisingly, their performance remains sensitive to perturbations that have little effect on human performance. In this work, we propose a novel extension of Mixup called Robustmix that regularizes networks to classify based on lower-frequency spatial features. We show that this type of regularization improves robustness on a range of benchmarks such as Imagenet-C and Stylized Imagenet. It adds little computational overhead and, furthermore, does not require a priori knowledge of a large set of image transformations. We find that this approach further complements recent advances in model architecture and data augmentation, attaining a state-of-the-art mCE of 44.8 with an EfficientNet-B8 model and RandAugment, which is a reduction of 16 mCE compared to the baseline. |
| title | Robustmix: Improving Robustness by Regularizing the Frequency Bias of Deep Nets |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2304.02847 |