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

As van Rooyen et al. noted in the first sentence of the abstract of [vRMW15], "Convex potential minimisation is the de facto approach to binary classification." Given the ubiquity of this approach, it is natural to study its abilities and limitations in the presence of noise, andindeed this isthe subject of many works (see e.g.