When Hardness of Approximation Meets Hardness of Learning