Robust Mixture Models for Algorithmic Fairness Under Latent Heterogeneity

2025-09-22

AI Summary

Standard machine learning models exhibit poor performance on latent subgroups—heterogeneous groups implicitly defined by complex interactions between continuous and discrete features—and lack robustness to distributional shifts. To address this, we propose ROME (Robust Optimization via Mixture Estimation), a framework that automatically discovers latent subgroup structure without requiring predefined group labels, while jointly optimizing both worst-group accuracy (group robustness) and overall average performance. ROME unifies modeling for both linear settings (via an EM algorithm) and nonlinear settings (via neural mixture-of-experts), effectively capturing high-order feature interactions. Empirical evaluation on synthetic and real-world benchmarks demonstrates that ROME consistently improves worst-group accuracy by 3.2–12.7 percentage points over strong baselines, while maintaining competitive average performance. This makes ROME particularly suitable for scenarios involving unknown or dynamically evolving sources of unfairness.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Mixed Discrete/Continuous Search
• Intelligent Robots: Learning & Optimization for ROB

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
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Standard machine learning models optimized for average performance often fail on minority subgroups and lack robustness to distribution shifts. This challenge worsens when subgroups are latent and affected by complex interactions among continuous and discrete features. We introduce ROME (RObust Mixture Ensemble), a framework that learns latent group structure from data while optimizing for worst-group performance. ROME employs two approaches: an Expectation-Maximization algorithm for linear models and a neural Mixture-of-Experts for nonlinear settings. Through simulations and experiments on real-world datasets, we demonstrate that ROME significantly improves algorithmic fairness compared to standard methods while maintaining competitive average performance. Importantly, our method requires no predefined group labels, making it practical when sources of disparities are unknown or evolving.

Problem

• Standard ML models fail on minority subgroups lacking robustness
• Latent subgroups affected by complex feature interactions worsen fairness
• Method requires no predefined group labels for unknown disparities

Innovation

• Learns latent group structure from data
• Optimizes for worst-group performance robustness
• Uses EM algorithm and neural Mixture-of-Experts