Version française / Séminaires
Séminaire MODAL'X : Vincent Divol (CEREMADE, Université Paris Dauphine - PSL)
Publié le 27 août 2024
–
Mis à jour le 2 octobre 2024
Fair regression in the unaware framework using optimal transport
Date(s)
le 3 octobre 2024
13h30-14h30
Lieu(x)
Résumé : Statistical fairness seeks to ensure an equitable distribution of predictions or algorithmic decisions across different sensitive groups. Among the fairness criteria under consideration, demographic parity is arguably the most conceptually straightforward: it simply requires that the distribution of outcomes is identical across all sensitive groups. We focus on the unawareness framework, where the prediction function cannot make direct use of the sensitive attribute (thus, individuals are not treated differently based on discriminating factors). In a regression setting, we solve the problem of finding the optimal unaware prediction function satisfying the demographic parity constraint. Interestingly enough, this problem boils down to solving a two-dimensional optimal transport problem for some degenerate cost. We study this optimal transport problem and exhibit relevant properties of the associated prediction function.
Mis à jour le 02 octobre 2024