Secure and Practical Federated Learning with Multi-Key Homomorphic Encryption

Nowadays, our data is stored and processed in the cloud, and there are increasing privacy concerns. Federated Learning (FL) has recently emerged to allow users to keep their data private and collaboratively train machine learning models. To ensure that no client data can be revealed, there are multiple challenges in designing a system to be used in real applications: (1) different types of heterogeneity (data, resource, infrastructure), (2) efficiency, and (3) scalability. In this project, we aim to design a secure and flexible FL system, SmartFL, with multi-key homomorphic encryption (MKHE) algorithms. From this project, we expect peer-reviewed top conference or journal papers and patents. We expect that by the end of the project, the working prototypes will be developed, which can be merged into HCL Canada Inc’s product in the future. The prototype will be demonstrated to the partner organization to show its scalability, efficiency, and practicality.

Faculty Supervisor:

Omid Ardakanian;Euijin Choo;Omid Ardakanian

Student:

Partner:

HCL Canada

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

University of Alberta

Program: