Trusted execution environments for federated learning

Purpose and goal

The project purpose is to develop and evaluate a pilot solution to mitigate the challenge with protecting and creating trusted executions of machine learning on local clients based on secure enclaves. The projects main goal is to develop a pilot implementation where secure enclaves are used to create traceable and secure federated machine learning and to systematically evaluate this implementation based on two classes of machine learning models to reduce risks associated with implementation of a production-ready solution for decentralized AI.

Expected results and effects

The main project deliverable consists of a pilot implementation with applications of the two main use-cases. Implementation supporting execution of federated machine learning where secure enclaves are used to increase security, correctness and trust in the system. Implementation supporting execution of select sensitive parts of federated machine learning within the enclave and parts outside. A systematic evaluation of the pilot implementation from a performance perspective . Dissemination of results in the form of a publicly available report, code and presented at seminar.

Planned approach and implementation

The project is divided into four main workpackages with associated milestones. WP1 - Technical evaluation of identified and widely used secure enclave technologies which have a potential application for federated machine learning. WP2 - Implementation of prototype for verification of identities on clients as well as implement FEDn clients that utilise secure enclave technology for the two main use-cases. WP3 - Systematic performance tests of the implemented solution for the two use-cases. WP4 - Result dissemination to the Swedish industry and research communities.

External links

Grundprojekt utifrån vilket vidareutveckling kommer ske. Projektlandningssida