Saddle point and duality in the optimization theory of convex set functions