SANER: Annotation-free Societal Attribute Neutralizer for Debiasing CLIP

2024/08/19 by Yusuke Hirota, Min-Hung Chen, Hirota, Yusuke +9 · 5 citations
Computer Science · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Hate Speech and Cyberbullying Detection

paper · pdf · doi:10.48550/arxiv.2408.10202

openalex publication_date 2024/08/19 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Large-scale vision-language models, such as CLIP, are known to contain societal bias regarding protected attributes (e.g., gender, age). This paper aims to address the problems of societal bias in CLIP. Although previous studies have proposed to debias societal bias through adversarial learning or test-time projecting, our comprehensive study of these works identifies two critical limitations: 1) loss of attribute information when it is explicitly disclosed in the input and 2) use of the attribute annotations during debiasing process. To mitigate societal bias in CLIP and overcome these limitations simultaneously, we introduce a simple-yet-effective debiasing method called SANER (societal attribute neutralizer) that eliminates attribute information from CLIP text features only of attribute-neutral descriptions. Experimental results show that SANER, which does not require attribute annotations and preserves original information for attribute-specific descriptions, demonstrates superior debiasing ability than the existing methods.

Cited by

• Aligned but Stereotypical? The Hidden Influence of System Prompts on Social Bias in LVLM-Based Text-to-Image Models
• Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models
• Person-Centric Annotations of LAION-400M: Auditing Bias and Its Transfer to Models
• Bias in Gender Bias Benchmarks: How Spurious Features Distort Evaluation
• LOTUS: A Leaderboard for Detailed Image Captioning from Quality to Societal Bias and User Preferences

Related

• Joint Vision-Language Social Bias Removal for CLIP
• DeAR: Debiasing Vision-Language Models with Additive Residuals
• Debiasing Large Vision-Language Models by Ablating Protected Attribute Representations
• Unlabeled Debiasing in Downstream Tasks via Class-wise Low Variance Regularization
• FairCLIP: Social Bias Elimination based on Attribute Prototype Learning and Representation Neutralization