Asymptotic optimality for active learning processes