Date of Defense
11-2025 1:00 PM
Location
H4-1006
Document Type
Thesis Defense
Degree Name
Master of Science in Information Security
College
College of Information Technology
Department
Information Security
First Advisor
Prof. Mohammad Mehedy Masud
Keywords
Federated Learning (FL), Homomorphic Encryption (HE), Privacy-Preserving Machine Learning, Efficiency, Model Updates Encryption
Abstract
Federated Learning (FL) is a decentralized approach of machine learning on multiple clients jointly training models without sharing their raw data, which drastically improves privacy and enhance protection against security breach. However, there is still a risk of privacy breach when clients send their model updates to the central server, because if a model update is intercepted or analyzed by a malicious entity, it could be used to recover sensitive data using inference attack. To address this issue, Homomorphic Encryption (HE) can be applied to protect against the interception, since the model updates remain encrypted during transmission as well as during aggregation. This is possible because Homomorphic Encryption allows computations to be performed directly on encrypted data without requiring decryption. In this research we propose an effective framework by integrating FL with HE. We focus on the encryption of model updates using HE prior to sending them to the central server. The server will perform aggregation on the encrypted updates using HE algorithms and send the encrypted aggregated model to the clients. Then the clients decrypt the aggregated model and again refine it using local data. After that, another cycle of model encryption, transmission to the central server and aggregation to the central server is conducted. Using this method, we intend to reduce the risk of leaking sensitive data and keep the efficiency of federated learning in an optimal level. We expect this study to result in (i) a robust privacy-preserving FL framework that aligns with IT management best practices for efficient and secure data handling. (ii) significant improvement in the level of potential privacy vulnerabilities, and (iii) assessment of the trade-offs between improving performance and computational overhead vs. maintaining privacy.
Included in
ENHANCED PRIVACY PRESERVING HEALTHCARE DATA MANAGEMENT WITH FEDERATED LEARNING USING HOMOMORPHIC ENCRYPTION
H4-1006
Federated Learning (FL) is a decentralized approach of machine learning on multiple clients jointly training models without sharing their raw data, which drastically improves privacy and enhance protection against security breach. However, there is still a risk of privacy breach when clients send their model updates to the central server, because if a model update is intercepted or analyzed by a malicious entity, it could be used to recover sensitive data using inference attack. To address this issue, Homomorphic Encryption (HE) can be applied to protect against the interception, since the model updates remain encrypted during transmission as well as during aggregation. This is possible because Homomorphic Encryption allows computations to be performed directly on encrypted data without requiring decryption. In this research we propose an effective framework by integrating FL with HE. We focus on the encryption of model updates using HE prior to sending them to the central server. The server will perform aggregation on the encrypted updates using HE algorithms and send the encrypted aggregated model to the clients. Then the clients decrypt the aggregated model and again refine it using local data. After that, another cycle of model encryption, transmission to the central server and aggregation to the central server is conducted. Using this method, we intend to reduce the risk of leaking sensitive data and keep the efficiency of federated learning in an optimal level. We expect this study to result in (i) a robust privacy-preserving FL framework that aligns with IT management best practices for efficient and secure data handling. (ii) significant improvement in the level of potential privacy vulnerabilities, and (iii) assessment of the trade-offs between improving performance and computational overhead vs. maintaining privacy.