Sample Complexity Bounds for Robustly Learning Decision Lists against Evasion Attacks. (arXiv:2205.06127v1 [cs.LG])

Web: http://arxiv.org/abs/2205.06127

A fundamental problem in adversarial machine learning is to quantify how much
training data is needed in the presence of evasion attacks. In this paper we
address this issue within the framework of PAC learning, focusing on the class
of decision lists. Given that distributional assumptions are essential in the
adversarial setting, we work with probability distributions on the input data
that satisfy a Lipschitz condition: nearby points have similar probability. Our
key results illustrate that the adversary's budget (that …

arxiv attacks complexity decision learning lists