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

2025-12-23

AI Summary

Diffusion models often amplify social biases—such as gender and skin-tone biases—present in training data during image synthesis, yet it remains unclear whether the sampling process itself modulates this bias amplification. This work presents the first systematic causal investigation into how sampling algorithms (e.g., DDIM, Euler) and hyperparameters (e.g., step count, eta) influence bias propagation. Leveraging controlled experiments on Biased MNIST, Multi-Color MNIST, BFFHQ, and Stable Diffusion—combined with quantitative bias metrics—we demonstrate that sampling strategies can actively mitigate or exacerbate bias, inducing bias magnitude shifts exceeding 40%. Crucially, these effects persist even when the underlying diffusion model is held fixed, revealing sampling as a controllable, post-training lever for bias regulation. Our findings challenge the prevailing assumption that bias stems solely from data or model architecture, and establish sampling as a novel, actionable intervention point for fair generative modeling.

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
• Responsible Web: Algorithmic accountability and transparency on the web

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_i_met_your_bias.

Problem

• Investigates bias amplification in diffusion models
• Analyzes sampling algorithms' impact on bias
• Demonstrates hyperparameters can reduce or amplify bias

Innovation

• Analyzes sampling algorithms' impact on bias amplification
• Shows hyperparameters can reduce or amplify dataset biases
• Empirical studies with controlled datasets and Stable Diffusion