“…However, large systems require large amounts of generated data, and most data adds little knowledge to the database. For instance, Jafar [37] uses the Latin hypercube sampling (LHS) approach to uniformly sample the entire search space, and researchers in [38] sample within the feasible neighbourhood of OCs, while researchers in [39] proposed an outer approximation to convexify the original nonconvex feasible space, then sample from the convex region to generate samples close to the security boundary. Venzke [40] uses infeasibility certificates based on separating hyperplanes to discard large portions of the input space as infeasible.…”