Conformal e-prediction in the presence of confounding

2026-03-11

AI Summary

This work addresses the challenge of maintaining valid coverage guarantees in conformal prediction when observational confounders distort the relationship between inputs and labels. It presents the first extension of the conformal e-prediction framework to settings with observed confounding, introducing a confounding adjustment mechanism that unifies the treatment of both independent and identically distributed (i.i.d.) data and non-i.i.d. structures—such as those arising from temporal or spatial dependencies. By incorporating this adjustment, the proposed method preserves theoretical coverage guarantees while substantially enhancing applicability and robustness in confounded environments. The approach establishes a new paradigm for uncertainty quantification in the presence of observable confounders, offering a principled solution for reliable predictive inference under realistic, complex data-generating processes.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty
• Constraint Satisfaction and Optimization: Other Foundations of Constraint Satisfaction

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

This note extends conformal e-prediction to cover the case where there is observed confounding between the random object $X$ and its label $Y$. We consider both the case where the observed data is IID and a case where some dependence between observations is permitted.

Problem

• conformal e-prediction
• confounding
• observed confounding
• IID data
• dependence

Innovation

• conformal e-prediction
• confounding
• dependence
• distribution-free inference