Excess-risk of distributed stochastic learners