Data Bias Mitigation under Coverage Constraints & The Price of Fairness
Researchers have developed a new framework to mitigate bias in machine learning models, particularly for individuals at the intersection of multiple sensitive attributes like race and gender. This approach incorporates coverage constraints to ensure sufficient representation across all subgroups, including intersectional ones. The method formulates bias mitigation as an integer linear program, optimizing strategies and quantifying the "price of fairness" as a function of tolerance, allowing for informed trade-offs between bias reduction and data modification costs. Evaluations on public datasets show that this framework preserves predictive accuracy and downstream ML performance. AI
IMPACT This research offers a novel approach to address complex bias issues in AI, potentially leading to fairer and more reliable ML systems across various applications.