Randomized learning: Generalization performance of old and new theoretically grounded algorithms