How I Met Your Bias: Investigating Bias Amplification in Diffusion Models

2025/12/23 by Nathan Roos, Roos, Nathan, Ekaterina Iakovleva +7 · 1 citation
Biochemistry, Genetics and Molecular Biology · Computer Science · Materials Science · #Cell Image Analysis Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Machine Learning (cs.LG) #Machine Learning in Materials Science

paper · doi:10.48550/arxiv.2512.20233

openalex publication_date 2025/12/23 · openalex created_date 2025/12/25 · openalex updated_date 2026/07/28

Abstract

Diffusion-based generative models demonstrate state-of-the-art performance across various image synthesis tasks, yet their tendency to replicate and amplify dataset biases remains poorly understood. Although previous research has viewed bias amplification as an inherent characteristic of diffusion models, this work provides the first analysis of how sampling algorithms and their hyperparameters influence bias amplification. We empirically demonstrate that samplers for diffusion models -- commonly optimized for sample quality and speed -- have a significant and measurable effect on bias amplification. Through controlled studies with models trained on Biased MNIST, Multi-Color MNIST and BFFHQ, and with Stable Diffusion, we show that sampling hyperparameters can induce both bias reduction and amplification, even when the trained model is fixed. Source code is available at https://github.com/How-I-met-your-bias/how_imet_your_bias.

Citations

• Frequency Regulation for Exposure Bias Mitigation in Diffusion Models
• Deeper Diffusion Models Amplify Bias
• Responsible Diffusion Models via Constraining Text Embeddings within Safe Regions
• Dissecting and Mitigating Diffusion Bias via Mechanistic Interpretability
• Diffusing DeBias: Synthetic Bias Amplification for Model Debiasing
• Generated Bias: Auditing Internal Bias Dynamics of Text-To-Image Generative Models
• Bias Begets Bias: The Impact of Biased Embeddings on Diffusion Models
• Gender Bias Evaluation in Text-to-image Generation: A Survey
• OpenBias: Open-set Bias Detection in Text-to-Image Generative Models
• Balancing Act: Distribution-Guided Debiasing in Diffusion Models
• Debiasing Text-to-Image Diffusion Models
• Quantifying Bias in Text-to-Image Generative Models
• DREAM: Diffusion Rectification and Estimation-Adaptive Models
• Unmasking Bias in Diffusion Model Training
• Improved Baselines with Visual Instruction Tuning
• Elucidating the Exposure Bias in Diffusion Models
• The Bias Amplification Paradox in Text-to-Image Generation
• Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis
• T2IAT: Measuring Valence and Stereotypical Biases in Text-to-Image Generation
• Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps
• Uncurated Image-Text Datasets: Shedding Light on Demographic Bias
• Debiasing Scores and Prompts of 2D Diffusion for View-consistent Text-to-3D Generation
• Input Perturbation Reduces Exposure Bias in Diffusion Models
• Discovering and Mitigating Visual Biases through Keyword Explanation
• LAION-5B: An open large-scale dataset for training next generation image-text models
• Classifier-Free Diffusion Guidance
• Discover and Mitigate Unknown Biases with Debiasing Alternate Networks
• DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps
• Elucidating the Design Space of Diffusion-Based Generative Models
• High-Resolution Image Synthesis with Latent Diffusion Models
• BiaSwap: Removing dataset bias with bias-tailored swapping augmentation
• Learning Debiased Representation via Disentangled Feature Augmentation
• Diffusion Models Beat GANs on Image Synthesis
• Learning Transferable Visual Models From Natural Language Supervision
• Zero-Shot Text-to-Image Generation
• Score-Based Generative Modeling through Stochastic Differential Equations
• Denoising Diffusion Implicit Models
• Denoising Diffusion Probabilistic Models
• Learning De-biased Representations with Biased Representations
• A Style-Based Generator Architecture for Generative Adversarial Networks
• U-Net: Convolutional Networks for Biomedical Image Segmentation
• THE USE OF CONFIDENCE OR FIDUCIAL LIMITS ILLUSTRATED IN THE CASE OF THE BINOMIAL

Cited by

• Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics

Related

• Deeper Diffusion Models Amplify Bias
• Unmasking Bias in Diffusion Model Training
• Exploiting the Signal-Leak Bias in Diffusion Models
• Elucidating the Exposure Bias in Diffusion Models
• Rethinking Noise Sampling in Class-Imbalanced Diffusion Models