Abstract:The most prevalent notions of fairness in machine learning are statistical definitions: they fix a small collection of pre-defined groups, and then ask for parity of some statistic of the classifier across these groups. Constraints of this form are susceptible to intentional or inadvertent "fairness gerrymandering", in which a classifier appears to be fair on each individual group, but badly violates the fairness constraint on one or more structured subgroups defined over the protected attributes. We propose instead to demand statistical notions of fairness across exponentially (or infinitely) many subgroups, defined by a structured class of functions over the protected attributes. This interpolates between statistical definitions of fairness and recently proposed individual notions of fairness, but raises several computational challenges. It is no longer clear how to audit a fixed classifier to see if it satisfies such a strong definition of fairness. We prove that the computational problem of auditing subgroup fairness for both equality of false positive rates and statistical parity is equivalent to the problem of weak agnostic learning, which means it is computationally hard in the worst case, even for simple structured subclasses. We then derive two algorithms that provably converge to the best fair classifier, given access to oracles which can solve the agnostic learning problem. The algorithms are based on a formulation of subgroup fairness as a two-player zero-sum game between a Learner and an Auditor. Our first algorithm provably converges in a polynomial number of steps. Our second algorithm enjoys only provably asymptotic convergence, but has the merit of simplicity and faster per-step computation. We implement the simpler algorithm using linear regression as a heuristic oracle, and show that we can effectively both audit and learn fair classifiers on real datasets.

To Pool or Not To Pool: Analyzing the Regularizing Effects of Group-Fair Training on Shared Models

Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness

An Empirical Study of Rich Subgroup Fairness for Machine Learning

On Learning Fairness and Accuracy on Multiple Subgroups

Bayes-Optimal Classifiers under Group Fairness

Fairness Through Equality of Effort

Fair Multi-party Machine Learning -- a Game Theoretic approach

Mitigating Group Bias in Federated Learning: Beyond Local Fairness

Metrizing Fairness

Learning by Grouping: A Multilevel Optimization Framework for Improving Fairness in Classification without Losing Accuracy

Federating for Learning Group Fair Models

Towards A Holistic View of Bias in Machine Learning: Bridging Algorithmic Fairness and Imbalanced Learning

Focus on the Common Good: Group Distributional Robustness Follows

Federated Fairness without Access to Sensitive Groups

Fairness Warnings and Fair-MAML: Learning Fairly with Minimal Data

On The Impact of Machine Learning Randomness on Group Fairness

Fairness in Machine Learning with Tractable Models

Robust Optimization for Fairness with Noisy Protected Groups

Fairness-Aware Class Imbalanced Learning on Multiple Subgroups

On the (In)Compatibility between Group Fairness and Individual Fairness