BiasMap: Leveraging Cross-Attentions to Discover and Mitigate Hidden Social Biases in Text-to-Image Generation

2025-09-16

AI Summary

This work addresses implicit concept-level social biases—e.g., gender or racial stereotypes—in text-to-image diffusion models (e.g., Stable Diffusion). We propose BiasMap, a model-agnostic framework for bias discovery and mitigation. Unlike prior methods that analyze only output distributions, BiasMap introduces cross-attention attribution maps to quantify the entanglement between sensitive attributes (e.g., gender, race) and semantic concepts (e.g., occupation) within the latent representation space, and defines SoftIoU as an interpretable coupling metric. Furthermore, we integrate energy-guided diffusion sampling to directly optimize the latent noise space during denoising, enforcing attribute–concept disentanglement. Experiments demonstrate that BiasMap significantly improves statistical fairness in generated images while substantially reducing spatial coupling between sensitive attributes and semantic concepts at the representation level—enhancing both transparency and controllability of the generation process.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Ethics, Bias, and Fairness
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Bias discovery is critical for black-box generative models, especiall text-to-image (TTI) models. Existing works predominantly focus on output-level demographic distributions, which do not neces- sarily guarantee concept representations to be disentangled post- mitigation. We propose BiasMap, a model-agnostic framework for uncovering latent concept-level representational biases in stable dif- fusion models. BiasMap leverages cross-attention attribution maps to reveal structural entanglements between demographics (e.g., gender, race) and semantics (e.g., professions), going deeper into representational bias during the image generation. Using attribu- tion maps of these concepts, we quantify the spatial demographics- semantics concept entanglement via Intersection over Union (IoU), offering a lens into bias that remains hidden in existing fairness dis- covery approaches. In addition, we further utilize BiasMap for bias mitigation through energy-guided diffusion sampling that directly modifies latent noise space and minimizes the expected SoftIoU dur- ing the denoising process. Our findings show that existing fairness interventions may reduce the output distributional gap but often fail to disentangle concept-level coupling, whereas our mitigation method can mitigate concept entanglement in image generation while complementing distributional bias mitigation.

Problem

• Discover hidden social biases in text-to-image generation models
• Quantify demographic-semantic entanglement using cross-attention maps
• Mitigate concept-level bias through energy-guided diffusion sampling

Innovation

• Cross-attention attribution maps reveal biases
• Intersection over Union quantifies concept entanglement
• Energy-guided diffusion sampling minimizes bias