Efficient Federated Conformal Prediction with Group-Conditional Guarantees

TL;DR

This paper introduces GC-FCP, a federated conformal prediction method that ensures group-conditional coverage guarantees by using mergeable, group-stratified coresets, enabling efficient and privacy-preserving uncertainty quantification across distributed data sources.

Contribution

The paper presents a novel federated conformal prediction protocol that provides group-conditional guarantees using mergeable coresets, addressing challenges in distributed, group-structured data.

Findings

GC-FCP achieves accurate group-conditional coverage in federated settings.

The method enables communication-efficient calibration through compact coresets.

Experimental results show GC-FCP outperforms centralized baselines in diverse datasets.

Abstract

Deploying trustworthy AI systems requires principled uncertainty quantification. Conformal prediction (CP) is a widely used framework for constructing prediction sets with distribution-free coverage guarantees. In many practical settings, including healthcare, finance, and mobile sensing, the calibration data required for CP are distributed across multiple clients, each with its own local data distribution. In this federated setting, data can often be partitioned into, potentially overlapping, groups, which may reflect client-specific strata or cross-cutting attributes such as demographic or semantic categories. We propose group-conditional federated conformal prediction (GC-FCP), a novel protocol that provides group-conditional coverage guarantees. GC-FCP constructs mergeable, group-stratified coresets from local calibration scores, enabling clients to communicate compact weighted summaries that support efficient aggregation and calibration at the server. Experiments on synthetic and real-world datasets validate the performance of GC-FCP compared to centralized calibration baselines.