Learning Non-Discriminatory Predictors
Abstract
We consider learning a predictor which is non-discriminatory with respect to a “protected attribute” according to the notion of “equalized odds” proposed by Hardt et al. (2016). We study the problem of learning such a non-discriminatory predictor from a finite training set, both statistically and computationally. We show that a post-hoc correction approach, as suggested by Hardt et al, can be highly suboptimal, present a nearly-optimal statistical procedure, argue that the associated computational problem is intractable, and suggest a second moment relaxation of the non-discrimination definition for which learning is tractable.
Cite
Text
Woodworth et al. "Learning Non-Discriminatory Predictors." Proceedings of the 2017 Conference on Learning Theory, 2017.Markdown
[Woodworth et al. "Learning Non-Discriminatory Predictors." Proceedings of the 2017 Conference on Learning Theory, 2017.](https://mlanthology.org/colt/2017/woodworth2017colt-learning/)BibTeX
@inproceedings{woodworth2017colt-learning,
title = {{Learning Non-Discriminatory Predictors}},
author = {Woodworth, Blake and Gunasekar, Suriya and Ohannessian, Mesrob I. and Srebro, Nathan},
booktitle = {Proceedings of the 2017 Conference on Learning Theory},
year = {2017},
pages = {1920-1953},
volume = {65},
url = {https://mlanthology.org/colt/2017/woodworth2017colt-learning/}
}