Faster Fair Machine via Transferring Fairness Constraints to Virtual Samples

Fair classification is an emerging and important research topic in machine learning community. Existing methods usually formulate the fairness metrics as additional inequality constraints, then embed them into original objective. This makes fair problems unable to be effectively tackled by some solvers specific unconstrained optimization. Although many new tailored algorithms have been designed attempt overcome this limitation, they often increase computation burden cannot cope with all types of metrics. To address these challenging issues, paper, we propose a novel method for classification. Specifically, theoretically demonstrate that linear non-linear covariance functions can transferred two virtual samples, which existing state-of-the-art applicable cases. Meanwhile, generalize proposed multiple constraints. We take SVM example show effectiveness our idea. Empirically, test on real-world datasets results confirm its excellent performance.