Collective Counterfactual Explanations Via Optimal Transport

Counterfactual explanations provide individuals with cost-optimal actionsthat can alter their labels to desired classes. However, if substantialinstances seek state modification, such individual-centric methods can lead tonew competitions and unanticipated costs. Furthermore, these recommendations,disregarding the underlying data distribution, may suggest actions that usersperceive as outliers. To address these issues, our work proposes a collectiveapproach for formulating counterfactual explanations, with an emphasis onutilizing the current density of the individuals to inform the recommendedactions. Our problem naturally casts as an optimal transport problem.Leveraging the extensive literature on optimal transport, we illustrate howthis collective method improves upon the desiderata of classical counterfactualexplanations. We support our proposal with numerical simulations, illustratingthe effectiveness of the proposed approach and its relation to classic methods.