On preserving non-discrimination when combining expert advice
arXiv:1810.11829
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.
arXiv commentAppeared in NIPS 2018
Cited by in corpus (7)
- Fairness of Exposure in Stochastic Bandits
- Fairness with Continuous Optimal Transport
- Towards Return Parity in Markov Decision Processes
- Beyond Individual and Group Fairness
- Evaluating Fairness of Machine Learning Models Under Uncertain and Incomplete Information
- Defining AI in Policy versus Practice
- Best-Case Lower Bounds in Online Learning