The vulnerability of learning to adversarial perturbation increases with intrinsic dimensionality | IEEE Conference Publication | IEEE Xplore