Controllable Feature Whitening for Hyperparameter-Free Bias Mitigation

2025-07-27

AI Summary

This work addresses bias in deep neural networks arising from spurious correlations learned from training data. We propose a controllable feature whitening framework that eliminates linear dependencies between target variables and sensitive (bias-inducing) features—without requiring regularization or adversarial training. Our method quantifies and removes such dependencies via a weighted covariance matrix, unifying demographic parity and equalized odds fairness criteria within a single, hyperparameter-free formulation that enables stable and tunable utility-fairness trade-offs. Unlike existing approaches that model higher-order dependencies or suffer from optimization instability, our framework relies solely on second-order covariance structure, ensuring computational efficiency and training robustness. Evaluated on four benchmarks—Corrupted CIFAR-10, Biased FFHQ, WaterBirds, and Celeb-A—our method significantly reduces bias while preserving or even improving predictive accuracy, demonstrating both effectiveness and strong generalization across diverse domains.

Technology Category

• Computer Vision: Bias, Fairness & Privacy
• Machine Learning: Ethics, Bias, and Fairness
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

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

Abstract

As the use of artificial intelligence rapidly increases, the development of trustworthy artificial intelligence has become important. However, recent studies have shown that deep neural networks are susceptible to learn spurious correlations present in datasets. To improve the reliability, we propose a simple yet effective framework called controllable feature whitening. We quantify the linear correlation between the target and bias features by the covariance matrix, and eliminate it through the whitening module. Our results systemically demonstrate that removing the linear correlations between features fed into the last linear classifier significantly mitigates the bias, while avoiding the need to model intractable higher-order dependencies. A particular advantage of the proposed method is that it does not require regularization terms or adversarial learning, which often leads to unstable optimization in practice. Furthermore, we show that two fairness criteria, demographic parity and equalized odds, can be effectively handled by whitening with the re-weighted covariance matrix. Consequently, our method controls the trade-off between the utility and fairness of algorithms by adjusting the weighting coefficient. Finally, we validate that our method outperforms existing approaches on four benchmark datasets: Corrupted CIFAR-10, Biased FFHQ, WaterBirds, and Celeb-A.

Problem

• Mitigate bias in deep neural networks
• Remove linear correlations between features
• Balance utility and fairness in algorithms

Innovation

• Controllable feature whitening for bias mitigation
• Whitening module eliminates linear feature correlations
• No regularization or adversarial learning required