On the influence of bias-correction on distributed stochastic optimization