Mirror descent and nonlinear projected subgradient methods for convex optimization