Images Speak Louder than Words: Understanding and Mitigating Bias in Vision-Language Model from a Causal Mediation Perspective

2024-07-03

Conference on Empirical Methods in Natural Language Processing

AI Summary

Vision-language models (VLMs) implicitly acquire gender bias during pretraining; existing debiasing methods primarily rely on input perturbation and output-level statistics, lacking mechanistic explanations of how bias emerges and propagates across internal model components. Method: This work introduces causal mediation analysis—the first such application in multimodal bias attribution—to construct an interpretable framework that precisely identifies bias sources and their interactions within the image encoder (contributing >32%), text encoder, and fusion module. We further propose component-level interventions and targeted blurring strategies. Results: Our approach reduces gender bias by 22.03% on MSCOCO and 9.04% on PASCAL-SENTENCE, with negligible performance degradation and no additional computational overhead.

Technology Category

• Computer Vision: Bias, Fairness & Privacy
• Machine Learning: Multimodal Learning
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Vision-language models (VLMs) pre-trained on extensive datasets can inadvertently learn biases by correlating gender information with specific objects or scenarios. Current methods, which focus on modifying inputs and monitoring changes in the model’s output probability scores, often struggle to comprehensively understand bias from the perspective of model components. We propose a framework that incorporates causal mediation analysis to measure and map the pathways of bias generation and propagation within VLMs. Our framework is applicable to a wide range of vision-language and multimodal tasks. In this work, we apply it to the object detection task and implement it on the GLIP model. This approach allows us to identify the direct effects of interventions on model bias and the indirect effects of interventions on bias mediated through different model components. Our results show that image features are the primary contributors to bias, with significantly higher impacts than text features, specifically accounting for 32.57% and 12.63% of the bias in the MSCOCO and PASCAL-SENTENCE datasets, respectively. Notably, the image encoder’s contribution surpasses that of the text encoder and the deep fusion encoder. Further experimentation confirms that contributions from both language and vision modalities are aligned and non-conflicting. Consequently, focusing on blurring gender representations within the image encoder which contributes most to the model bias, reduces bias efficiently by 22.03% and 9.04% in the MSCOCO and PASCAL-SENTENCE datasets, respectively, with minimal performance loss or increased computational demands.

Problem

• Measure and mitigate bias in vision-language models using causal mediation analysis
• Identify image features as primary bias contributors in VLMs
• Reduce bias by targeting gender representations in image encoder

Innovation

• Uses causal mediation analysis for bias pathways
• Focuses on image encoder for bias reduction
• Blurs gender representations to mitigate bias