The paper proposes a mini-batched versions of matrix stochastic gradient and regularized matric stochastic gradient for PCA. It presents two algorithms based on a convex relaxation to the PCA problem, with convergence guarantees for both of them. Numerical results could be improved.