Conformalized Neural Networks for Federated Uncertainty Quantification under Dual Heterogeneity

Researchers propose FedWQ-CP, a new approach for uncertainty quantification in federated learning that addresses both data and model heterogeneity challenges. The method enables reliable uncertainty estimation across distributed agents while maintaining efficiency through single-round communication and weighted threshold aggregation.

• →FedWQ-CP addresses dual heterogeneity challenges in federated learning that existing approaches handle separately.
• →The method performs agent-server calibration in just one communication round, improving efficiency.
• →Each agent computes local quantile thresholds which are aggregated through weighted averaging at the server.
• →Testing on seven public datasets shows the approach maintains coverage while producing smaller prediction sets.
• →The solution helps prevent overconfident model deployment at under-resourced federated learning agents.