🤖 AI Summary
This work investigates how enforcing individual fairness affects group fairness—a critical yet underexplored tension in fair machine learning.
Method: We propose a generalized entropy (GE)-based framework for quantifying and constraining individual fairness, and design an empirical risk minimization (ERM) classifier incorporating GE fairness constraints. Using VC-dimension analysis, we establish, for the first time, its PAC learnability.
Results: We theoretically and empirically demonstrate that strengthening individual fairness does not necessarily improve—nor even preserve—group fairness; excessive individual-level constraints can degrade group-level fairness metrics. This reveals a counterintuitive but fundamental trade-off boundary in fairness-aware learning. Extensive experiments on multiple benchmark datasets validate the theoretical findings, providing both a novel paradigm and rigorous theoretical foundations for navigating fairness trade-offs.
📝 Abstract
This paper investigates how the degree of group fairness changes when the degree of individual fairness is actively controlled. As a metric quantifying individual fairness, we consider generalized entropy (GE) recently introduced into machine learning community. To control the degree of individual fairness, we design a classification algorithm satisfying a given degree of individual fairness through an empirical risk minimization (ERM) with a fairness constraint specified in terms of GE. We show the PAC learnability of the fair ERM problem by proving that the true fairness degree does not deviate much from an empirical one with high probability for finite VC dimension if the sample size is big enough. Our experiments show that strengthening individual fairness degree does not always lead to enhancement of group fairness.
