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

ACL ARR 2024 June Submission2951 Authors

Abstract: 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.

Paper Type: Long

Research Area: Ethics, Bias, and Fairness

Research Area Keywords: Ethics, Fairness, Gender bias, Generative model, Dataset-Level Bias Mitigation

Contribution Types: Model analysis & interpretability

Languages Studied: English

Submission Number: 2951