SLIDE 1 Obtaining Fairness using Optimal Transport Theory
Authors: Eustasio del Barrio, Fabrice Gamboa, Paula Gordaliza , Jean-Michel Loubes Presenter: Theo Hu
SLIDE 2 Fairness Definition in this paper
Demographic Parity: Basically says that your prediction should be independent of sensitive attribute S · Disparate Impact(DI) · Balanced Error Rate(BER)
SLIDE 3
Total Variation Distance is a good way to bound unfairness
SLIDE 4
Two general ways of removing ‘unfairness’
SLIDE 5
Total Repair and Wasserstein barycenter
SLIDE 6
Total Repair and Wasserstein barycenter
SLIDE 7
Computing Barycenter
SLIDE 8
Bound of Utility Lost
SLIDE 9
Total vs. Partial Repair
SLIDE 10
Random Repair and its guarantee
SLIDE 11
Comparison to Geometric Repair (Previous work)
SLIDE 12
Computational Aspects
SLIDE 13
Computational Aspects- Total Repair
SLIDE 14
Computational Aspects- Random Repair
SLIDE 15
Experiments
SLIDE 16
Case that Random Repair beats Geometric Repair
