On preserving non-discrimination when combining expert advice

Abstract

We study the interplay between sequential decision making and avoiding discrimination against protected groups, when examples arrive online and do not follow distributional assumptions. We consider the most basic extension of classical online learning: Given a class of predictors that are individually non-discriminatory with respect to a particular metric, how can we combine them to perform as well as the best predictor, while preserving non-discrimination? Surprisingly we show that this task is unachievable for the prevalent notion of equalized odds that requires equal false negative rates and equal false positive rates across groups. On the positive side, for another notion of non-discrimination, equalized error rates, we show that running separate instances of the classical multiplicative weights algorithm for each group achieves this guarantee. Interestingly, even for this notion, we show that algorithms with stronger performance guarantees than multiplicative weights cannot preserve non-discrimination.

Online Fair Revenue Maximizing Cake Division with Non-Contiguous Pieces in Adversarial Bandits

“…These works propose statistical criteria to test algorithmic fairness that sometimes exploits definitions of fairness from political philosophy and sociology. Several prior works like Blum et al, 2018 [11] and Blum & Lykouris, 2019 [12] study how to achieve these fairness criteria in online learning. These algorithms achieve fairness to the incoming users like what we do here.…”