Hyper-parameter Tuning for Adversarially Robust Models