Trusted execution environments for federated learning

Purpose and goal:

The project´s goal has been to evaluate the possibility of enabling privacy-preserving machine learning with minimal effort from the end user on hardware that guarantees both data integrity and application integrity, thereby enabling collaborations that would otherwise not have been possible.

This goal has been achieved by developing open software that complements the existing suite of open software for federated machine learning. Users of FEDn can now evaluate Trusted Execution Environments (TEEs) as an option for additional security and integrity guarantees. Expected results and effects:

The results are made available as open source on Scaleout´s GitHub page. A technical report is also published.

We expect that the project´s results can be used for educational purposes, where users and prospective customers can review the technical report being prepared, and also use the software that has been developed. Additionally, we expect that we can use the methodology developed in customer projects where we need to guarantee data integrity and ensure that models are trained exactly according to specification.

Approach and implementation: Phase 1: Preparation. Planning and investigation of appropriate technology choices.

Phase 2: Implementation. Development of the software required to run all the necessary components of a TEE. Development of attestation service.

Phase 3: Benchmarking and report writing. A careful examination of the limitations of this technology in terms of performance and memory, and which data and models are appropriate. Authoring of technical report.