Learning with incremental iterative regularization