This paper introduces a formal framework for "performative prediction with selective labels," addressing scenarios where machine learning models influence population behavior and labels are only observed for a subset of the population. The research demonstrates that standard retraining methods can fail to converge to a stable solution when labels are selective. To counter this, the authors propose a worst-case objective using confidence intervals on label probabilities, which, when combined with repeated risk minimization, ensures the model stays close to the true stable point. Experiments in a lending application with fairness regularization show this robust optimization approach achieves performance comparable to methods using complete label access. AI
IMPACT Introduces a novel approach to handle selective labels in performative prediction, potentially improving fairness and robustness in real-world ML applications.
RANK_REASON This is a research paper detailing a new theoretical framework and experimental results in machine learning. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- fairness regularization
- Hugging Face
- lending application
- machine learning
- Performative Prediction
- Performative Prediction with Selective Labels
- Repeated Risk Minimization
- RNA recognition motif domain
- robust optimization
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →