Equity-Regularized Federated Multimodal Learning with Missingness-Aware Gating and Subgroup- Conditional Conformal Risk Control for Clinical Prediction
Abstract
Equity-Regularized Federated Multimodal Learning with Missingness-Aware Gating and Subgroup-Conditional Conformal Risk Control for Clinical Prediction presents PRISM-Fed, a privacy-preserving federated learning framework designed to improve the fairness, robustness, and reliability of clinical prediction across heterogeneous healthcare institutions. The framework integrates a Missingness-Aware Gating (MAG) mechanism to effectively handle incomplete multimodal clinical data without requiring imputation, an Equity-Regularized Federated Aggregation (EqFedAvg) strategy that prioritizes underperforming patient subgroups during model optimization, and Subgroup-Conditional Conformal Risk Control (SC-CRC) to provide distribution-free, subgroup-specific risk guarantees with calibrated clinical decision thresholds. By jointly addressing missing modalities, demographic disparities, and uncertainty calibration within a federated environment, PRISM-Fed enables trustworthy, equitable, and privacy-preserving clinical prediction while maintaining high predictive performance and robustness under real-world deployment conditions.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.