Adaptive Adversarial Training Does Not Increase Recourse Costs

Abstract

Recent work has connected adversarial attack methods and algorithmic recourse methods: both seek minimal changes to an input instance which alter a model's classication decision. It has been shown that traditional adversarial training, which seeks to minimize a classier's susceptibility to malicious perturbations, increases the cost of generated recourse; with larger adversarial training radii correlating with higher recourse costs. From the perspective of algorithmic recourse, however, the appropriate adversarial training radius has always been unknown. Another recent line of work has motivated adversarial training with adaptive training radii to address the issue of instance-wise variable adversarial vulnerability, showing success in domains with unknown attack radii. This work studies the eects of adaptive adversarial training on algorithmic recourse costs. We establish that the improvements in model robustness induced by adaptive adversarial training show little eect on algorithmic recourse costs, providing a potential avenue for aordable robustness in domains where recoursability is critical. CCS CONCEPTS• Theory of computation ! Adversarial learning; • Computing methodologies ! Knowledge representation and reasoning; • Human-centered computing ! Human computer interaction (HCI).