The perils of being unhinged: On the accuracy of classifiers minimizing a noise-robust convex loss

Van Rooyen et al. introduced a notion of convex loss functions being robust\nto random classification noise, and established that the "unhinged" loss\nfunction is robust in this sense. In this note we study the accuracy of binary\nclassifiers obtained by minimizing the unhinged loss, and observe that even for\nsimple linearly separable data distributions, minimizing the unhinged loss may\nonly yield a binary classifier with accuracy no better than random guessing.\n