Virginia Tech
Practical Privacy-Preserving Federated Learning with Secure Multi-Party Computation
Abstract
dc:description.abstractFederated Learning (FL) has become increasingly important for machine learning applica- tions that do not allow user data to be removed from the device it exists on. Basic FL is still susceptible to attacks that compromise privacy through privacy inference and poisoning attacks. Through secure Multi-Party Computation (MPC) and Differential Privacy (DP) it is possible to eliminate more of these attacks. Secure Multi-Party Computation is a promising efficient solution for privacy-preserving com- putation to ensure that the aggregation servers do not learn the information being protected. The implementation of MPC has occurred in many different formats and security threats. Prior work SAFEFL utilizes the use of MPC through MP-SPDZ, a secure MPC framework, in their secure federated learning framework, however, it suffered from numerous limitations in execution time that prevented any practical use. In this study, SAFEFL's limitations are examined thoroughly and discussed how to be im- proved. A new design is proposed for secure federated learning with MPC. Additionally, differential privacy is added into the MPC aggregation for an extra layer of security. These new designs are compared with the original SAFEFL design and show the performance improvements and scalability for future.
Degree
thesis:*- Name thesis:degree_name
- Master of Science
- Level thesis:degree_level
- masters
- Discipline thesis:degree_discipline
- Computer Engineering
- Department dc:contributor.department
- Electrical and Computer Engineering
- Grantor dc:publisher
- Virginia Tech
- Year dc:date.issued
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Akhtar, Benjamin Asad
- Chair dc:contributor.committeechair
-
- Xiong, Wenjie
- Committee members dc:contributor.committeemember
-
- Shao, Linbo
- Hoang, Thang
Subjects
dc:subject × 4Rights
dc:rights- Statement dc:rights
-
- In Copyright
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Other
- vt_gsexam:41285