Related Experiment Videos
Privacy-preserving clustered federated learning via differential privacy and homomorphically encrypted prototypes
Jun Zhan1, Zhenglong Jiang1, Lang Liu2
1School of Information Engineering, Jingdezhen University, Jingdezhen, 333400, China.
Scientific Reports
|July 21, 2026
Summary
Privacy-Preserving Clustered Federated Learning (PPCFL) enhances data privacy by protecting clustering signals and model updates. This novel approach improves accuracy in non-IID settings while safeguarding sensitive client information.
Area of Science:
- Machine Learning
- Distributed Systems
- Cryptography
Background:
- Clustered Federated Learning (CFL) addresses statistical heterogeneity by grouping clients.
- Existing CFL methods risk exposing sensitive clustering information and updates to the server.
- This can reveal client similarities and compromise privacy.
Purpose of the Study:
- To propose a novel Privacy-Preserving Clustered Federated Learning (PPCFL) framework.
- To enhance privacy protection in CFL by securing clustering signals and cluster-specific updates.
- To improve the privacy-utility trade-off in dynamic Non-IID settings.
Main Methods:
- A split-stream framework integrating adaptive Gaussian perturbation and threshold Paillier encrypted aggregation.
- Backbone updates use adaptive Gaussian perturbation; clustering signatures and head updates use stream-specific perturbation and Paillier encryption.
- Server aggregates in ciphertext domain; decryption is performed by a client subset, preventing server decryption.
Main Results:
- PPCFL achieves the highest final-round accuracy across MNIST, Fashion-MNIST, and CIFAR-10 datasets.
- Demonstrates enhanced protection for clustering information and cluster-specific updates.
- Improves final accuracy over DP-FedAvg and IFCA under Dirichlet settings.
Conclusions:
- PPCFL effectively balances privacy and utility in clustered federated learning.
- The proposed methods offer robust protection against privacy leakage of clustering-related information.
- PPCFL represents a significant advancement for privacy-preserving machine learning in heterogeneous environments.