Tag
This paper investigates the limits of proper learning and regularization in multiclass learning, resolving open problems by demonstrating that learning cannot always be reduced to proper learning and that regularization has structural constraints.