Researchers have developed a closed-form linear projector that can reduce the influence of protected attributes in two-sided matching systems at inference time. This method aims to ensure fairness by bounding the welfare change caused by such edits. A pre-deployment screen, combining a verticality test and a welfare sweep, categorizes markets as PASS, KILL, or OUT-OF-SCOPE. The system was tested on synthetic and realistic markets, showing that architecture choices, such as using two-tower or graph neural network encoders, can impact the admissibility of post-hoc edits. AI
IMPACT This research could lead to more robust fairness guarantees in AI systems used for resource allocation and decision-making.
RANK_REASON The item is a research paper detailing a new method for screening fairness edits in AI systems. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- encoder retraining
- graph neural network
- Hugging Face
- linear projector
- Probabilistic Refinement of Model-Based Segmentation: Application to Radiation Therapy Planning of the Head and Neck
- two-sided matching markets
- When Is a Cheap Fairness Edit Safe? Screening Post-Hoc Edits in Two-Sided Matching Markets
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
