Fairness Through Computationally-Bounded Awareness
This work addresses fairness in machine learning for classification tasks, but it is incremental as it builds on existing metric-based frameworks with a bounded query constraint.
The paper tackles the problem of fair classification by introducing a new fairness notion called metric multifairness, which ensures similar subpopulations are treated similarly under a bounded query model to an arbitrary metric, and shows how to achieve this notion in their setting.
We study the problem of fair classification within the versatile framework of Dwork et al. [ITCS '12], which assumes the existence of a metric that measures similarity between pairs of individuals. Unlike earlier work, we do not assume that the entire metric is known to the learning algorithm; instead, the learner can query this arbitrary metric a bounded number of times. We propose a new notion of fairness called metric multifairness and show how to achieve this notion in our setting. Metric multifairness is parameterized by a similarity metric $d$ on pairs of individuals to classify and a rich collection ${\cal C}$ of (possibly overlapping) "comparison sets" over pairs of individuals. At a high level, metric multifairness guarantees that similar subpopulations are treated similarly, as long as these subpopulations are identified within the class ${\cal C}$.