A paper showing that the convergence rate of proper loss, which is commonly used in classification is governed by the modulus of convexity of the generalized entropy function has been accepted by the Conference on Learning Theory (COLT).

Han Bao.
Proper Losses, Moduli of Convexity, and Surrogate Regret Bounds.
In Proceedings of the 36th Conference on Learning Theory (COLT), 2023.