Fair Conformal Classification via Learning Representation-Based Groups

2026-05-12

AI Summary

This work proposes a fair conformal classification framework that integrates representation learning with conformal prediction to address the limitation of traditional conformal methods, which only guarantee marginal coverage and often fail to ensure equitable coverage across subgroups. By implicitly learning nonlinear feature interactions, the framework adaptively identifies vulnerable subpopulations and provides them with conditional coverage guarantees. The approach maintains the compactness and validity of prediction sets while significantly improving coverage fairness at the subgroup level. Experimental results on multiple synthetic and real-world datasets demonstrate that the method effectively detects undercovered subgroups and enhances coverage equity, thereby increasing both the reliability and fairness of predictive models.

Abstract

Conformal prediction methods provide statistically rigorous marginal coverage guarantees for machine learning models, but such guarantees fail to account for algorithmic biases, thereby undermining fairness and trust. This paper introduces a fair conformal inference framework for classification tasks. The proposed method constructs prediction sets that guarantee conditional coverage on adaptively identified subgroups, which can be implicitly defined through nonlinear feature combinations. By balancing effectiveness and efficiency in producing compact, informative prediction sets and ensuring adaptive equalized coverage across unfairly treated subgroups, our approach paves a practical pathway toward trustworthy machine learning. Extensive experiments on both synthetic and real-world datasets demonstrate the effectiveness of the framework.

Problem

• conformal prediction
• fairness
• conditional coverage
• algorithmic bias
• subgroup fairness

Innovation

• Fair Conformal Prediction
• Representation-Based Subgroups
• Conditional Coverage
• Adaptive Equalized Coverage
• Nonlinear Feature Combinations