Post-processing fairness with minimal changes
This work addresses fairness in AI for applications where minimal disruption to existing predictions is critical, though it is incremental as it builds on existing post-processing methods.
The paper tackles the problem of post-processing fairness in machine learning by introducing a model-agnostic algorithm that enforces minimal changes between biased and debiased predictions without requiring sensitive attributes at test time. The result is demonstrated through empirical evaluations, showing competitive performance against four other debiasing algorithms on two widely used datasets.
In this paper, we introduce a novel post-processing algorithm that is both model-agnostic and does not require the sensitive attribute at test time. In addition, our algorithm is explicitly designed to enforce minimal changes between biased and debiased predictions; a property that, while highly desirable, is rarely prioritized as an explicit objective in fairness literature. Our approach leverages a multiplicative factor applied to the logit value of probability scores produced by a black-box classifier. We demonstrate the efficacy of our method through empirical evaluations, comparing its performance against other four debiasing algorithms on two widely used datasets in fairness research.