SocialCounterfactuals: Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examples

While vision-language models (VLMs) have achieved re-markable performance improvements recently, there is growing evidence that these models also posses harmful biases with respect to social attributes such as gender and race. Prior studies have primarily focused on probing such bias attributes indi...

Full description

Saved in:
Bibliographic Details
Published inProceedings (IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Online) pp. 11975 - 11985
Main Authors Howard, Phillip, Madasu, Avinash, Le, Tiep, Moreno, Gustavo Lujan, Bhiwandiwalla, Anahita, Lal, Vasudev
Format Conference Proceeding
LanguageEnglish
Published IEEE 16.06.2024
Subjects
Computer vision
counterfactuals
Diffusion models
Fairness
intersectionality
Pattern recognition
Prevention and mitigation
social bias
Text to image
Training
vision-language models
Online AccessGet full text

Cover

More Information
Summary:While vision-language models (VLMs) have achieved re-markable performance improvements recently, there is growing evidence that these models also posses harmful biases with respect to social attributes such as gender and race. Prior studies have primarily focused on probing such bias attributes individually while ignoring biases associated with intersections between social attributes. This could be due to the difficulty of collecting an exhaustive set of imagetext pairs for various combinations of social attributes. To address this challenge, we employ text-to-image diffusion models to produce counterfactual examples for probing intersectional social biases at scale. Our approach utilizes Stable Diffusion with cross attention control to produce sets of counterfactual image-text pairs that are highly similar in their depiction of a subject (e.g., a given occupation) while differing only in their depiction of intersectional social attributes (e.g., race & gender). Through our over-generate-then-filter methodology, we produce SocialCounterfactuals, a high-quality dataset containing 171k image-text pairs for probing intersectional biases related to gender, race, and physical characteristics. We conduct extensive experiments to demonstrate the usefulness of our generated dataset for probing and mitigating intersectional social biases in state-of-the-art VLMs.
ISSN:1063-6919
DOI:10.1109/CVPR52733.2024.01138