GBMix: Enhancing Fairness by Group-Balanced Mixup
Mixup is a powerful data augmentation strategy that has been shown to improve the generalization and adversarial robustness of machine learning classifiers, particularly in computer vision applications. Despite its simplicity and effectiveness, the impact of Mixup on the fairness of a model has not...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2024-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10413562/ |
| _version_ | 1827355275529551872 |
|---|---|
| author | Sangwoo Hong Youngseok Yoon Hyungjun Joo Jungwoo Lee |
| author_facet | Sangwoo Hong Youngseok Yoon Hyungjun Joo Jungwoo Lee |
| author_sort | Sangwoo Hong |
| collection | DOAJ |
| description | Mixup is a powerful data augmentation strategy that has been shown to improve the generalization and adversarial robustness of machine learning classifiers, particularly in computer vision applications. Despite its simplicity and effectiveness, the impact of Mixup on the fairness of a model has not been thoroughly investigated yet. In this paper, we demonstrate that Mixup can perpetuate or even exacerbate bias presented in the training set. We provide insight to understand the reasons behind this behavior and propose GBMix, a group-balanced Mixup strategy to train fair classifiers. It groups the dataset based on their attributes and balances the Mixup ratio between the groups. Through the reorganization and balance of Mixup among groups, GBMix effectively enhances both average and worst-case accuracy concurrently. We empirically show that GBMix effectively mitigates bias in the training set and reduces the performance gap between groups. This effect is observed across a range of datasets and networks, and GBMix outperforms all the state-of-the-art methods. |
| first_indexed | 2024-03-08T04:08:20Z |
| format | Article |
| id | doaj.art-2e77c5614055405392aab5bbf121068f |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-03-08T04:08:20Z |
| publishDate | 2024-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-2e77c5614055405392aab5bbf121068f2024-02-09T00:03:24ZengIEEEIEEE Access2169-35362024-01-0112188771888710.1109/ACCESS.2024.335827510413562GBMix: Enhancing Fairness by Group-Balanced MixupSangwoo Hong0https://orcid.org/0000-0002-0270-2781Youngseok Yoon1Hyungjun Joo2Jungwoo Lee3https://orcid.org/0000-0002-6804-980XDepartment of Electrical and Computer Engineering, Communications and Machine Learning Laboratory, Seoul National University, Seoul, South KoreaDepartment of Electrical and Computer Engineering, Communications and Machine Learning Laboratory, Seoul National University, Seoul, South KoreaDepartment of Electrical and Computer Engineering, Communications and Machine Learning Laboratory, Seoul National University, Seoul, South KoreaDepartment of Electrical and Computer Engineering, Communications and Machine Learning Laboratory, Seoul National University, Seoul, South KoreaMixup is a powerful data augmentation strategy that has been shown to improve the generalization and adversarial robustness of machine learning classifiers, particularly in computer vision applications. Despite its simplicity and effectiveness, the impact of Mixup on the fairness of a model has not been thoroughly investigated yet. In this paper, we demonstrate that Mixup can perpetuate or even exacerbate bias presented in the training set. We provide insight to understand the reasons behind this behavior and propose GBMix, a group-balanced Mixup strategy to train fair classifiers. It groups the dataset based on their attributes and balances the Mixup ratio between the groups. Through the reorganization and balance of Mixup among groups, GBMix effectively enhances both average and worst-case accuracy concurrently. We empirically show that GBMix effectively mitigates bias in the training set and reduces the performance gap between groups. This effect is observed across a range of datasets and networks, and GBMix outperforms all the state-of-the-art methods.https://ieeexplore.ieee.org/document/10413562/Mixupfairnessdata augmentationbiasspurious correlation |
| spellingShingle | Sangwoo Hong Youngseok Yoon Hyungjun Joo Jungwoo Lee GBMix: Enhancing Fairness by Group-Balanced Mixup IEEE Access Mixup fairness data augmentation bias spurious correlation |
| title | GBMix: Enhancing Fairness by Group-Balanced Mixup |
| title_full | GBMix: Enhancing Fairness by Group-Balanced Mixup |
| title_fullStr | GBMix: Enhancing Fairness by Group-Balanced Mixup |
| title_full_unstemmed | GBMix: Enhancing Fairness by Group-Balanced Mixup |
| title_short | GBMix: Enhancing Fairness by Group-Balanced Mixup |
| title_sort | gbmix enhancing fairness by group balanced mixup |
| topic | Mixup fairness data augmentation bias spurious correlation |
| url | https://ieeexplore.ieee.org/document/10413562/ |
| work_keys_str_mv | AT sangwoohong gbmixenhancingfairnessbygroupbalancedmixup AT youngseokyoon gbmixenhancingfairnessbygroupbalancedmixup AT hyungjunjoo gbmixenhancingfairnessbygroupbalancedmixup AT jungwoolee gbmixenhancingfairnessbygroupbalancedmixup |