Adversarially Robust PAC Learning with Optimal VC Rates
Higher bounds of adversarial robustness are proved for every concept class of $\operatorname{VC}$ of $\operatorname{VC}$ dimension $d$ in the realizable and agnostic settings and an optimal first-order refinement of the latter, together with an optimal first-order refinement of the latter.
Steve Hanneke, Amirreza Shaeiri
· 0 citations