Privacy-preserving clustered federated learning via differential privacy and homomorphically encrypted prototypes
- 1School of Information Engineering, Jingdezhen University
- 2Cybersecurity and Information Technology Center, Jingdezhen University
Abstract
Clustered federated learning (CFL) is an effective paradigm for handling statistical heterogeneity by grouping clients with similar data characteristics and learning cluster-specific models. However, existing CFL methods often expose sensitive clustering signals or cluster-specific updates to the server, which may reveal latent client similarity relations and weaken privacy protection. To address this issue, we propose Privacy-Preserving Clustered Federated Learning (PPCFL), a split-stream framework that integrates adaptive Gaussian perturbation with threshold Paillier encrypted aggregation. In PPCFL, backbone updates are protected by adaptive Gaussian perturbation before plaintext aggregation, while clustering signatures and cluster-head updates are first perturbed by stream-specific adaptive Gaussian mechanisms and then uploaded under threshold Paillier encryption. The server performs ciphertext-domain aggregation for clustering prototypes and cluster-head updates, whereas plaintext prototypes and cluster-level decrypted aggregates are recovered by a qualified threshold-decryption client subset without giving the server decryption capability. In addition, PPCFL adopts round-wise budget growth, utility-aware refinement, and adaptive clipping-threshold updates to improve the privacy–utility trade-off under dynamic Non-IID settings. Experiments on MNIST, Fashion-MNIST, and CIFAR-10 show that PPCFL achieves the highest final-round accuracy among the evaluated methods in the reported settings while providing enhanced protection for clustering-related information and cluster-specific updates. Under the representative Dirichlet setting α =0.5 , PPCFL improves the final accuracy over DP-FedAvg by 0.33, 1.73, and 2.62 percentage points on MNIST, Fashion-MNIST, and CIFAR-10, respectively, and over IFCA by 0.98, 8.28, and 10.24 percentage points.
Related
- FFT-GA-BC: a fused federated transformer genetic algorithm framework with blockchain-secured, differentially private, homomorphically encrypted aggregation for mutation, phenotype, and binary-allele analysis.
- Federated learning for privacy-preserving opportunistic screening of lumbar-spine DXA-defined osteoporotic-level bone mineral density using chest radiographs: A multi-center independent validation study
- Montage-agnostic federated learning for privacy-preserving Alzheimer’s disease classification from heterogeneous multi-site EEG
- A privacy-preserving multicenter framework for lung nodule segmentation in low-dose CT using an F1-score-weighted federated 3D Transformer-UNet
- Mapping Aptamer Prototypes to the Unmet Gaps of Glioblastoma Care: From Circulating Cells to the Resection Cavity
BibTeX
@article{zhan2026privacypreserving,
title = {{Privacy-preserving clustered federated learning via differential privacy and homomorphically encrypted prototypes}},
author = {Jun Zhan and Zhenglong Jiang and Lang Liu},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-63242-3},
url = {https://doi.org/10.1038/s41598-026-63242-3},
note = {CC BY},
}
RIS
TY - JOUR
TI - Privacy-preserving clustered federated learning via differential privacy and homomorphically encrypted prototypes
AU - Jun Zhan
AU - Zhenglong Jiang
AU - Lang Liu
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-63242-3
UR - https://doi.org/10.1038/s41598-026-63242-3
AB - Clustered federated learning (CFL) is an effective paradigm for handling statistical heterogeneity by grouping clients with similar data characteristics and learning cluster-specific models. However, existing CFL methods often expose sensitive clustering signals or cluster-specific updates to the server, which may reveal latent client similarity relations and weaken privacy protection. To address this issue, we propose Privacy-Preserving Clustered Federated Learning (PPCFL), a split-stream framework that integrates adaptive Gaussian perturbation with threshold Paillier encrypted aggregation. In PPCFL, backbone updates are protected by adaptive Gaussian perturbation before plaintext aggregation, while clustering signatures and cluster-head updates are first perturbed by stream-specific adaptive Gaussian mechanisms and then uploaded under threshold Paillier encryption. The server performs ciphertext-domain aggregation for clustering prototypes and cluster-head updates, whereas plaintext prototypes and cluster-level decrypted aggregates are recovered by a qualified threshold-decryption client subset without giving the server decryption capability. In addition, PPCFL adopts round-wise budget growth, utility-aware refinement, and adaptive clipping-threshold updates to improve the privacy–utility trade-off under dynamic Non-IID settings. Experiments on MNIST, Fashion-MNIST, and CIFAR-10 show that PPCFL achieves the highest final-round accuracy among the evaluated methods in the reported settings while providing enhanced protection for clustering-related information and cluster-specific updates. Under the representative Dirichlet setting α =0.5 , PPCFL improves the final accuracy over DP-FedAvg by 0.33, 1.73, and 2.62 percentage points on MNIST, Fashion-MNIST, and CIFAR-10, respectively, and over IFCA by 0.98, 8.28, and 10.24 percentage points.
SN - 2045-2322
ER -