Reproduce “Fair Classification by Direct Intervention on Operating Characteristics” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kevin Jiang, Edgar Dobriban
University of Pennsylvania researchers developed ROCF, a post-processing framework for binary classification that addresses multiple linear fractional group fairness constraints by directly optimizing operating characteristics. This method effectively enforces approximate fairness for metrics like Demographic Parity, Equalized Odds, Predictive Parity, and False Omission Rate Parity, achieving near-oracle accuracy with minimal interventions on diverse datasets.
Key insights
- Reformulating the fairness problem into the lower-dimensional space of operating characteristics and their convex hulls enables direct and efficient optimization for complex fairness constraints.
- Linearizing non-convex linear fractional constraints through a centroid-based bi-level optimization strategy converts them into tractable linear programs.
- Carefully designed post-processing interventions can achieve approximate multi-constraint fairness with minimal accuracy degradation and few label modifications, making it suitable for practical deployment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.