Optimized first-order methods for smooth convex minimization