CAD-VAE: Leveraging Correlation-Aware Latents for Comprehensive Fair Disentanglement

2025-03-11

AI Summary

Deep generative models often exacerbate fairness biases due to natural correlations between sensitive attributes and target variables; enforcing strict disentanglement typically incurs information loss or is infeasible. To address this, we propose CAD-VAE (Correlation-Aware Disentangled VAE), the first VAE framework that explicitly models shared structural dependencies via a correlation-aware latent variable and achieves unsupervised separation of overlapping factors through conditional mutual information minimization. We further introduce correlation-driven optimization and contrastive reconstruction regularization to jointly preserve representational fidelity and eliminate redundancy. Crucially, CAD-VAE requires no domain-specific prior knowledge and enables controllable, interpretable fair disentanglement. Extensive experiments on multiple benchmark datasets demonstrate significant improvements in representation fairness (ΔDP ↓37%), high-fidelity counterfactual generation, and fine-grained, fairness-constrained image editing.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

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

Abstract

While deep generative models have significantly advanced representation learning, they may inherit or amplify biases and fairness issues by encoding sensitive attributes alongside predictive features. Enforcing strict independence in disentanglement is often unrealistic when target and sensitive factors are naturally correlated. To address this challenge, we propose CAD-VAE (Correlation-Aware Disentangled VAE), which introduces a correlated latent code to capture the shared information between target and sensitive attributes. Given this correlated latent, our method effectively separates overlapping factors without extra domain knowledge by directly minimizing the conditional mutual information between target and sensitive codes. A relevance-driven optimization strategy refines the correlated code by efficiently capturing essential correlated features and eliminating redundancy. Extensive experiments on benchmark datasets demonstrate that CAD-VAE produces fairer representations, realistic counterfactuals, and improved fairness-aware image editing.

Problem

• Addresses bias and fairness in deep generative models
• Separates correlated target and sensitive attributes
• Improves fairness-aware image editing and representations

Innovation

• Introduces correlated latent code for shared information
• Minimizes conditional mutual information for separation
• Uses relevance-driven optimization to refine features