Individual Fairness Guarantees for Neural Networks

Conference Paper (2022)
Author(s)

Elias Benussi (University of Oxford)

Andrea Patane (University of Oxford)

Matthew Wicker (University of Oxford)

Luca Laurenti (TU Delft - Team Luca Laurenti)

Marta Kwiatkowska (University of Oxford)

Research Group
Team Luca Laurenti
Copyright
© 2022 Elias Benussi, Andrea Patane, Matthew Wicker, L. Laurenti, Marta Kwiatkowska
DOI related publication
https://doi.org/10.24963/ijcai.2022/92
More Info
expand_more
Publication Year
2022
Language
English
Copyright
© 2022 Elias Benussi, Andrea Patane, Matthew Wicker, L. Laurenti, Marta Kwiatkowska
Research Group
Team Luca Laurenti
Pages (from-to)
651-658
ISBN (electronic)
978-1-956792-00-3
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

We consider the problem of certifying the individual fairness (IF) of feed-forward neural networks (NNs). In particular, we work with the ϵ-δ-IF formulation, which, given a NN and a similarity metric learnt from data, requires that the output difference between any pair of ϵ-similar individuals is bounded by a maximum decision tolerance δ ≥ 0. Working with a range of metrics, including the Mahalanobis distance, we propose a method to over-approximate the resulting optimisation problem using piecewise-linear functions to lower and upper bound the NN's non-linearities globally over the input space. We encode this computation as the solution of a Mixed-Integer Linear Programming problem and demonstrate that it can be used to compute IF guarantees on four datasets widely used for fairness benchmarking. We show how this formulation can be used to encourage models' fairness at training time by modifying the NN loss, and empirically confirm our approach yields NNs that are orders of magnitude fairer than state-of-the-art methods.

Files

0092.pdf
(pdf | 1.69 Mb)
License info not available