Equalized Generative Treatment: Matching f-divergences for Fairness in Generative Models

2026/02/09 by Alexandre Vérine, Alexandre Verine, Rafaël Pinot +2 · 1 voice
Computer Science · Medicine · Social Sciences · #Artificial Intelligence in Healthcare and Education #Computational and Text Analysis Methods #Generative Adversarial Networks and Image Synthesis #Generative grammar #Generative model #Group (periodic table) #Matching (statistics) #Quality (philosophy) #Set (abstract data type) #cs.AI #cs.LG

paper · pdf · doi:10.48550/arxiv.2602.08660

openalex publication_date 2026/02/09 · arxiv published 2026/02/09 · arxiv updated 2026/02/09 · openalex created_date 2026/02/11 · openalex updated_date 2026/07/28

Abstract

Fairness is a crucial concern for generative models, which not only reflect but can also amplify societal and cultural biases. Existing fairness notions for generative models are largely adapted from classification and focus on balancing the probability of generating samples from each sensitive group. We show that such criteria are brittle, as they can be met even when different sensitive groups are modeled with widely varying quality. To address this limitation, we introduce a new fairness definition for generative models, termed as equalized generative treatment (EGT), which requires comparable generation quality across all sensitive groups, with quality measured via a reference f-divergence. We further analyze the trade-offs induced by EGT, demonstrating that enforcing fairness constraints necessarily couples the overall model quality to that of the most challenging group to approximate. This indicates that a simple yet efficient min-max fine-tuning method should be able to balance f-divergences across sensitive groups to satisfy EGT. We validate this theoretical insight through a set of experiments on both image and text generation tasks. We demonstrate that min-max methods consistently achieve fairer outcomes compared to other approaches from the literature, while maintaining competitive overall performance for both tasks.

Citations

• Improving Diversity in Language Models: When Temperature Fails, Change the Loss
• Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models
• Simulating the Real World: A Unified Survey of Multimodal Generative Models
• One-step Diffusion Models with f-Divergence Distribution Matching
• Fair Bilevel Neural Network (FairBiNN): On Balancing fairness and accuracy via Stackelberg Equilibrium
• PFGuard: A Generative Framework with Privacy and Fairness Safeguards
• Inverse-RLignment: Large Language Model Alignment from Demonstrations through Inverse Reinforcement Learning
• Improving Fairness and Mitigating MADness in Generative Models
• Training Unbiased Diffusion Models From Biased Dataset
• Balancing Act: Distribution-Guided Debiasing in Diffusion Models
• Exploring Precision and Recall to assess the quality and diversity of LLMs
• Beyond Reverse KL: Generalizing Direct Preference Optimization with Diverse Divergence Constraints
• Are aligned neural networks adversarially aligned?
• TopP&R: Robust Support Estimation Approach for Evaluating Fidelity and Diversity in Generative Models
• DINOv2: Learning Robust Visual Features without Supervision
• Aligning Language Models with Preferences through f-divergence Minimization
• Extracting Training Data from Diffusion Models
• MAUVE Scores for Generative Models: Theory and Practice
• Fair Generative Models via Transfer Learning
• Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models
• Elucidating the Design Space of Diffusion-Based Generative Models
• Quantifying Memorization Across Neural Language Models
• FORML: Learning to Reweight Data for Fairness
• MaGNET: Uniform Sampling from Deep Generative Network Manifolds Without Retraining
• Generative Adversarial Networks
• Measuring Fairness in Generative Models
• Pros and Cons of GAN Evaluation Measures: New Developments
• Maximum Likelihood Training of Score-Based Diffusion Models
• Analyzing and Improving the Image Quality of StyleGAN
• Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization
• Fair Generative Modeling via Weak Supervision
• Improved Precision and Recall Metric for Assessing Generative Models
• A Style-Based Generator Architecture for Generative Adversarial Networks
• A Style-Based Generator Architecture for Generative Adversarial Networks
• 50 Years of Test (Un)fairness: Lessons for Machine Learning
• Assessing Generative Models via Precision and Recall
• Texygen: A Benchmarking Platform for Text Generation Models
• How (not) to Train your Generative Model: Scheduled Sampling, Likelihood, Adversary?
• Variational Inference with Normalizing Flows

Discussions

• 📖 New LAMSADE preprint: Equalized Generative Treatment: Matching f-divergences for Fairness in Generative Models. (arXiv) https://doi.org/10.48550/arXiv.2602.08660 [bsky, 0 points, 0 comments]

Related

• Fair Diffusion: Instructing Text-to-Image Generation Models on Fairness
• On the Fairness, Diversity and Reliability of Text-to-Image Generative Models
• AIM-Fair: Advancing Algorithmic Fairness via Selectively Fine-Tuning Biased Models with Contextual Synthetic Data
• On Measuring Fairness in Generative Models
• Measuring Fairness in Generative Models