Stav dette: Pseudo-derandomizing learning and approximation