Machine learning (ML) models are vulnerable to evasion attacks, where the attacker adds almost imperceptible perturbation to a correctly classified instance so as to induce misclassification. In the black-box setting where ...