Addressing Fairness in Classification with a Model-Agnostic Multi-Objective Algorithm

Kirtan Padh, Diego Antognini, Emma Lejal-Glaude, Boi Faltings, Claudiu Musat

Abstract:

The goal of fairness in classification is to learn a
classifier that does not discriminate against groups
of individuals based on sensitive attributes, such as
race and gender. One approach to designing fair al-
gorithms is to use relaxations of fairness notions as
regularization terms or in a constrained optimiza-
tion problem. We observe that the hyperbolic tan-
gent function can approximate the indicator func-
tion. We leverage this property to define a differen-
tiable relaxation that approximates fairness notions
provably better than existing relaxations. In addi-
tion, we propose a model-agnostic multi-objective
architecture that can simultaneously optimize for
multiple fairness notions and multiple sensitive
attributes and supports all statistical parity-based
notions of fairness. We use our relaxation with
the multi-objective architecture to learn fair clas-
sifiers. Experiments on public datasets show that
our method suffers a significantly lower loss of ac-
curacy than current debiasing algorithms relative
to the unconstrained model.