Resampled Datasets Are Not Enough: Mitigating Societal Bias Beyond Single Attributes

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hirota, Yusuke, Andrews, Jerone T. A., Zhao, Dora, Papakyriakopoulos, Orestis, Modas, Apostolos, Nakashima, Yuta, Xiang, Alice
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929417413984256
author Hirota, Yusuke
Andrews, Jerone T. A.
Zhao, Dora
Papakyriakopoulos, Orestis
Modas, Apostolos
Nakashima, Yuta
Xiang, Alice
author_facet Hirota, Yusuke
Andrews, Jerone T. A.
Zhao, Dora
Papakyriakopoulos, Orestis
Modas, Apostolos
Nakashima, Yuta
Xiang, Alice
contents We tackle societal bias in image-text datasets by removing spurious correlations between protected groups and image attributes. Traditional methods only target labeled attributes, ignoring biases from unlabeled ones. Using text-guided inpainting models, our approach ensures protected group independence from all attributes and mitigates inpainting biases through data filtering. Evaluations on multi-label image classification and image captioning tasks show our method effectively reduces bias without compromising performance across various models.
format Preprint
id arxiv_https___arxiv_org_abs_2407_03623
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Resampled Datasets Are Not Enough: Mitigating Societal Bias Beyond Single Attributes
Hirota, Yusuke
Andrews, Jerone T. A.
Zhao, Dora
Papakyriakopoulos, Orestis
Modas, Apostolos
Nakashima, Yuta
Xiang, Alice
Computer Vision and Pattern Recognition
We tackle societal bias in image-text datasets by removing spurious correlations between protected groups and image attributes. Traditional methods only target labeled attributes, ignoring biases from unlabeled ones. Using text-guided inpainting models, our approach ensures protected group independence from all attributes and mitigates inpainting biases through data filtering. Evaluations on multi-label image classification and image captioning tasks show our method effectively reduces bias without compromising performance across various models.
title Resampled Datasets Are Not Enough: Mitigating Societal Bias Beyond Single Attributes
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2407.03623