A Platform for Privacy-preserving Machine Learning Using the Ethereum Blockchain and Smart Contracts
|Coordinator||Scaleout Systems AB|
|Funding from Vinnova||SEK 1 693 803|
|Project duration||August 2019 - September 2020|
|Venture||Collaboration projects in cybersecurity and digital infrastructure|
|Call||Cybersecurity and reliability - Theme: Identity and block chains|
Purpose and goal
The purpose of the project was to develop and demonstrate the possibilities for privacy-preserving federated machine learning with blockchain components. This goal has been achieved through the development of the project solution. The project has aimed to promote privacy-preserving AI collaborations and it has been achieved by fulfilling the following three sub-goals, an application suite for federated machine learning with blockchain checkpointing, governance using smart contracts and tools for defining incentive models and creating markets.
Expected results and effects
The results are made available through open source code (STACKn + FEDn) and demonstrators describing two relevant applications. Furthermore, the results have contributed to membership in AI Sweden´s partner network and the solution is used in the test bed project to explore federated user cases. The results of the project will be applied by participating in the EU project AICHAIN for the agency Eurocontrol with, among others, SwissAir as a participant. We commercialize through expansion of the project results and packaging of applications to solve, for example, data silo problems.
Planned approach and implementation
The project has been divided into three phases where phase 1 consisted of project establishment and preparation. Project phase 2 consisted of development increments where the reference group has been involved in the presentation of the results. Phase 3 of the project consisted of demonstrations of developed applications for predictive maintenance and digital pathology. To ensure the project results are supported, the reference group has been continuously involved.