Fair Logistic Regression: An Adversarial Perspective
release_6moo2j4fjfbozd4saj3di5tpyi
by
Ashkan Rezaei, Rizal Fathony, Omid Memarrast, Brian Ziebart
2019
Abstract
Fair prediction methods have primarily been built around existing
classification techniques using pre-processing methods, post-hoc adjustments,
reduction-based constructions, or deep learning procedures. We investigate a
new approach to fair data-driven decision making by designing predictors with
fairness requirements integrated into their core formulations. We augment a
game-theoretic construction of the logistic regression model with fairness
constraints, producing a novel prediction model that robustly and fairly
minimizes the logarithmic loss. We demonstrate the advantages of our approach
on a range of benchmark datasets for fairness.
In text/plain
format
Archived Files and Locations
application/pdf 671.8 kB
file_3trc6v7nbvbnbjtfihdhbjiu24
|
arxiv.org (repository) web.archive.org (webarchive) |
1903.03910v2
access all versions, variants, and formats of this works (eg, pre-prints)