Extending Prediction-Powered Inference through Conformal Prediction

2025-10-17

AI Summary

This paper addresses the robustness of statistical parameter inference for black-box machine learning models under privacy constraints, distributional shift, and missing data. We propose the first method integrating conformal prediction into the Prediction-Augmented Inference (PAI) framework: it employs a calibration-set predictor for imputation and uniformly supports inference for means, M-estimators, and e-values. Crucially, we introduce the first offline-computable, general-purpose e-value PAI construction. The method is inherently compatible with differential privacy, robust to covariate shift, and model-agnostic—requiring no access to internal model structure. Experiments on private datasets and time-series data demonstrate significant improvements in both inferential validity and privacy-preserving security. Our approach establishes a new paradigm for trustworthy statistical inference in complex, dynamic environments.

Technology Category

• Machine Learning: Privacy
• Reasoning under Uncertainty: Probabilistic Inference
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Security and Privacy: Security and privacy of machine learning and AI applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Prediction-powered inference is a recent methodology for the safe use of black-box ML models to impute missing data, strengthening inference of statistical parameters. However, many applications require strong properties besides valid inference, such as privacy, robustness or validity under continuous distribution shifts; deriving prediction-powered methods with such guarantees is generally an arduous process, and has to be done case by case. In this paper, we resolve this issue by connecting prediction-powered inference with conformal prediction: by performing imputation through a calibrated conformal set-predictor, we attain validity while achieving additional guarantees in a natural manner. We instantiate our procedure for the inference of means, Z- and M-estimation, as well as e-values and e-value-based procedures. Furthermore, in the case of e-values, ours is the first general prediction-powered procedure that operates off-line. We demonstrate these advantages by applying our method on private and time-series data. Both tasks are nontrivial within the standard prediction-powered framework but become natural under our method.

Problem

• Extending prediction-powered inference with conformal guarantees
• Achieving validity under privacy and distribution shift constraints
• Providing general offline inference for e-values and estimation

Innovation

• Connects prediction-powered inference with conformal prediction
• Uses calibrated conformal set-predictor for imputation
• Achieves validity with privacy and robustness guarantees