Beyond Data Splitting: Full-Data Conformal Prediction by Differential Privacy

2026-03-08

AI Summary

This work proposes the first conformal prediction framework that achieves differential privacy without data splitting, thereby preserving sample efficiency while simultaneously ensuring valid coverage and producing tight prediction sets. Existing privacy-preserving conformal methods rely on data splitting, which compromises statistical efficiency and struggles to balance differential privacy, coverage guarantees, and set compactness. The proposed approach leverages differential privacy–induced stability to control the discrepancy between in-sample and out-of-sample conformity scores and introduces a conservative private quantile mechanism to prevent undercoverage. Theoretical analysis demonstrates that the method asymptotically recovers the nominal coverage level under differential privacy constraints. Empirical evaluations confirm that, compared to split-based baselines, the framework yields significantly tighter prediction sets while maintaining valid coverage.

Technology Category

• Machine Learning: Privacy
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Computer Vision: Bias, Fairness & Privacy

Application Category

• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Privacy protection and uncertainty quantification are increasingly important in data-driven decision making. Conformal prediction provides finite-sample marginal coverage, but existing private approaches often rely on data splitting, reducing the effective sample size. We propose a full-data privacy-preserving conformal prediction framework that avoids splitting. Our framework leverages stability induced by differential privacy to control the gap between in-sample and out-of-sample conformal scores, and pairs this with a conservative private quantile routine designed to prevent under-coverage. We show that a generic differential privacy guarantee yields a universal coverage floor, yet cannot generally recover the nominal $1-\alpha$ level. We then provide a refined, mechanism-specific stability analysis and yields asymptotic recovery of the nominal level. Experiments demonstrate sharper prediction sets than the split-based private baseline.

Problem

• conformal prediction
• differential privacy
• uncertainty quantification
• privacy protection
• data splitting

Innovation

• conformal prediction
• differential privacy
• full-data utilization
• coverage guarantee
• private quantile