Privacy-Utility Tradeoffs in Hierarchical Federated Learning Under Combined Differential Privacy and CKKS Homomorphic Encryption for Distributed Healthcare Networks
Abstract
Abstract Privacy-preserving federated learning for healthcare requires mechanisms that address both gradient privacy and transmission confidentiality. Existing approaches apply Differential Privacy (DP) and Homomorphic Encryption (HE) in isolation, leaving their combined tradeoff unquantified in Hierarchical Federated Learning (HFL) settings. This work presents an HFL framework integrating client-level \(((\epsilon, \delta))\)-DP with institution-level Cheon-Kim-Kim-Song (CKKS) HE, evaluated under privacy budgets \((\epsilon \in \{2, 4, 6, 8\})\) across MNIST and MIT-BIH Arrhythmia. DP introduces bounded accuracy loss ranging from \((1.25%)\) (\((\varepsilon = 8)\)) to \((1.88%)\) (\((\varepsilon = 2)\)) on MNIST and \((6.15%)\) to \((12.23%)\) on MIT-BIH Arrhythmia. CKKS HE introduces zero measurable accuracy loss across all evaluated configurations, isolating DP noise as the sole source of utility degradation and establishing a clear empirical separation between privacy and confidentiality costs. These results are reproducible across three independent random seeds, with accuracy standard deviations below \((0.13%)\) for MNIST and \((1.31%)\) for MIT-BIH Arrhythmia, providing empirical foundations for privacy-preserving federated learning deployment in healthcare.
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.