Adversarial and Random Transformations for Robust Domain Adaptation and Generalization
Data augmentation has been widely used to improve generalization in training deep neural networks. Recent works show that using worst-case transformations or adversarial augmentation strategies can significantly improve accuracy and robustness. However, due to the non-differentiable properties of im...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-06-01
|
| Series: | Sensors |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1424-8220/23/11/5273 |
| _version_ | 1797596683261968384 |
|---|---|
| author | Liang Xiao Jiaolong Xu Dawei Zhao Erke Shang Qi Zhu Bin Dai |
| author_facet | Liang Xiao Jiaolong Xu Dawei Zhao Erke Shang Qi Zhu Bin Dai |
| author_sort | Liang Xiao |
| collection | DOAJ |
| description | Data augmentation has been widely used to improve generalization in training deep neural networks. Recent works show that using worst-case transformations or adversarial augmentation strategies can significantly improve accuracy and robustness. However, due to the non-differentiable properties of image transformations, searching algorithms such as reinforcement learning or evolution strategy have to be applied, which are not computationally practical for large-scale problems. In this work, we show that by simply applying consistency training with random data augmentation, state-of-the-art results on domain adaptation (DA) and generalization (DG) can be obtained. To further improve the accuracy and robustness with adversarial examples, we propose a differentiable adversarial data augmentation method based on spatial transformer networks (STNs). The combined adversarial and random-transformation-based method outperforms the state-of-the-art on multiple DA and DG benchmark datasets. Furthermore, the proposed method shows desirable robustness to corruption, which is also validated on commonly used datasets. |
| first_indexed | 2024-03-11T02:56:29Z |
| format | Article |
| id | doaj.art-38851932cd0440d788b09ce8661e0a6a |
| institution | Directory Open Access Journal |
| issn | 1424-8220 |
| language | English |
| last_indexed | 2024-03-11T02:56:29Z |
| publishDate | 2023-06-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Sensors |
| spelling | doaj.art-38851932cd0440d788b09ce8661e0a6a2023-11-18T08:34:55ZengMDPI AGSensors1424-82202023-06-012311527310.3390/s23115273Adversarial and Random Transformations for Robust Domain Adaptation and GeneralizationLiang Xiao0Jiaolong Xu1Dawei Zhao2Erke Shang3Qi Zhu4Bin Dai5Unmanned Systems Technology Research Center, Defense Innovation Institute, Beijing 100071, ChinaUnmanned Systems Technology Research Center, Defense Innovation Institute, Beijing 100071, ChinaUnmanned Systems Technology Research Center, Defense Innovation Institute, Beijing 100071, ChinaUnmanned Systems Technology Research Center, Defense Innovation Institute, Beijing 100071, ChinaUnmanned Systems Technology Research Center, Defense Innovation Institute, Beijing 100071, ChinaUnmanned Systems Technology Research Center, Defense Innovation Institute, Beijing 100071, ChinaData augmentation has been widely used to improve generalization in training deep neural networks. Recent works show that using worst-case transformations or adversarial augmentation strategies can significantly improve accuracy and robustness. However, due to the non-differentiable properties of image transformations, searching algorithms such as reinforcement learning or evolution strategy have to be applied, which are not computationally practical for large-scale problems. In this work, we show that by simply applying consistency training with random data augmentation, state-of-the-art results on domain adaptation (DA) and generalization (DG) can be obtained. To further improve the accuracy and robustness with adversarial examples, we propose a differentiable adversarial data augmentation method based on spatial transformer networks (STNs). The combined adversarial and random-transformation-based method outperforms the state-of-the-art on multiple DA and DG benchmark datasets. Furthermore, the proposed method shows desirable robustness to corruption, which is also validated on commonly used datasets.https://www.mdpi.com/1424-8220/23/11/5273image classificationdomain adaptationdomain generalizationconsistency trainingspatial transformer networksadversarial transformations |
| spellingShingle | Liang Xiao Jiaolong Xu Dawei Zhao Erke Shang Qi Zhu Bin Dai Adversarial and Random Transformations for Robust Domain Adaptation and Generalization Sensors image classification domain adaptation domain generalization consistency training spatial transformer networks adversarial transformations |
| title | Adversarial and Random Transformations for Robust Domain Adaptation and Generalization |
| title_full | Adversarial and Random Transformations for Robust Domain Adaptation and Generalization |
| title_fullStr | Adversarial and Random Transformations for Robust Domain Adaptation and Generalization |
| title_full_unstemmed | Adversarial and Random Transformations for Robust Domain Adaptation and Generalization |
| title_short | Adversarial and Random Transformations for Robust Domain Adaptation and Generalization |
| title_sort | adversarial and random transformations for robust domain adaptation and generalization |
| topic | image classification domain adaptation domain generalization consistency training spatial transformer networks adversarial transformations |
| url | https://www.mdpi.com/1424-8220/23/11/5273 |
| work_keys_str_mv | AT liangxiao adversarialandrandomtransformationsforrobustdomainadaptationandgeneralization AT jiaolongxu adversarialandrandomtransformationsforrobustdomainadaptationandgeneralization AT daweizhao adversarialandrandomtransformationsforrobustdomainadaptationandgeneralization AT erkeshang adversarialandrandomtransformationsforrobustdomainadaptationandgeneralization AT qizhu adversarialandrandomtransformationsforrobustdomainadaptationandgeneralization AT bindai adversarialandrandomtransformationsforrobustdomainadaptationandgeneralization |