Zaslat SMS: Pseudo-derandomizing learning and approximation