In this paper, we consider the problem of Gaussian process (GP) optimization
with an added robustness requirement: The returned point may be perturbed by an
adversary, and we require the function value to remain as high as possible even
after this perturbation. This problem is motivate