Отправить по sms: Pseudo-derandomizing learning and approximation