Learning NonDiscriminatory Predictors
[edit]
Proceedings of the 2017 Conference on Learning Theory, PMLR 65:19201953, 2017.
Abstract
We consider learning a predictor which is nondiscriminatory 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 nondiscriminatory predictor from a finite training set, both statistically and computationally. We show that a posthoc correction approach, as suggested by Hardt et al, can be highly suboptimal, present a nearlyoptimal statistical procedure, argue that the associated computational problem is intractable, and suggest a second moment relaxation of the nondiscrimination definition for which learning is tractable.
Related Material


