Efficient fair principal component analysis

It has been shown that dimension reduction methods such as Principal Component Analysis (PCA) may be inherently prone to unfairness and treat data from different sensitive groups race, color, sex, etc., unfairly. In pursuit of fairness-enhancing dimensionality reduction, using the notion Pareto optimality, we propose an adaptive first-order algorithm learn a subspace preserves fairness, while slightly compromising reconstruction loss. Theoretically, provide sufficient conditions solution proposed belongs frontier for all groups; thereby, optimal trade-off between overall loss fairness constraints is guaranteed. We also convergence analysis our show its efficacy through empirical studies on datasets, which demonstrates superior performance in comparison with state-of-the-art algorithms. The fairness-aware PCA can efficiently generalized multiple group features effectively reduce decisions downstream tasks classification.