Citation

Sen, Anik (2026) Hybrid privacy-preserving federated learning framework with homomorphic encryption for secure data sharing. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

Secure data sharing has become a critical challenge in data-driven systems operating under strict privacy constraints, where sensitive information cannot be directly exchanged among collaborating entities. Federated Learning (FL) has emerged as a distributed learning paradigm that enables collaborative model training while keeping raw data local to participating clients, thereby supporting privacy-preserving data analysis. However, despite its decentralised architecture, FL remains vulnerable to information leakage through model updates and incurs significant computational overhead when cryptographic techniques are incorporated to strengthen privacy guarantees. This study proposes a hybrid privacy-preserving framework for secure data sharing in FL by integrating cryptographic protection mechanisms. The proposed framework employs Cheon-Kim-Kim-Song (CKKS)-based homomorphic encryption (HE) to enable secure computation over encrypted model updates and utilises secure multi-party computation (SMPC)-based additive masking techniques derived to ensure secure aggregation during the model update process. In addition, quantisation is applied to model updates prior to encryption in order to reduce numerical precision, thereby significantly lowering cryptographic computational cost and communication overhead associated with encrypted aggregation. The effectiveness of the proposed framework is evaluated through three experimental case studies, including diabetes diagnosis using artificial neural networks and FL integrating HE, intensive care unit (ICU) mortality prediction using quantised encrypted aggregation, and brain tumour classification using computation-aware federated optimisation. Experimental results demonstrate that the proposed hybrid framework enables efficient and secure data sharing across distributed clients while substantially reducing encryption-related computational overhead. Furthermore, the integration of quantisation and secure aggregation mechanisms maintains strong predictive capability of the global model without significant degradation in global model accuracy. These findings indicate that combining CKKSbased HE, additive masking-based secure aggregation, and quantisation-based model compression provides a scalable and practical approach for secure data sharing in FL environments operating under strict privacy constraints.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: QA76.9.A25 S46 2026
Uncontrolled Keywords:	Data encryption (Computer science)
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	05 Nov 2026 13:49
Last Modified:	05 Nov 2026 13:49
URII:	http://shdl.mmu.edu.my/id/eprint/17075

Downloads

Edit (login required)