Stochastic gradient descent algorithms for strongly convex functions at O(1/T) convergence rates