Exacerbating Algorithmic Bias through Fairness Attacks

Algorithmic fairness has attracted significant attention in recent years, with many quantitative measures suggested for characterizing the of different machine learning algorithms. Despite this interest, robustness those respect to an intentional adversarial attack not been properly addressed. Indeed, most focused on impact malicious attacks accuracy system, without any regard system's fairness. We propose new types data poisoning where adversary intentionally targets a system. Specifically, we two families that target measures. In anchoring attack, skew decision boundary by placing poisoned points near specific bias outcome. influence fairness, aim maximize covariance between sensitive attributes and outcome affect model. conduct extensive experiments indicate effectiveness our proposed attacks.