Գրեք սա: Pseudo-derandomizing learning and approximation