The Exclusion-Inclusion Gap for Evaluating Finite Pareto Approximations in Multiobjective Stochastic Optimization
We introduce the exclusion–inclusion gap for evaluating finite Pareto approximations constructed from stochastic objective estimates. The gap measures the maximum of two objective-space errors associated with excluding Pareto points from the returned set’s true upper image and including dominated points in the returned set’s true lower image. The terms in its approximation–estimation error decomposition admit … Read more