Title: FedCF: Fair Federated Conformal Prediction

Original: arXiv:2509.22907v2 Announce Type: replace

Abstract: Conformal Prediction (CP) is a prevalent methodology for assessing uncertainty within machine learning models. While conventional CP provides probabilistic assurances regarding the coverage of actual labels, it generally disregards sensitive attributes present in the data. Recent research has attempted to integrate fairness into CP by establishing conditional coverage guarantees for various subgroups. One notable approach is Conformal Fairness (CF). This study expands the CF framework to encompass Federated Learning, outlining a method to evaluate federated models for fairness through an examination of demographic disparities. We substantiate our approach with empirical experiments across multiple datasets from diverse domains, making full use of the exchangeability assumption.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC